DATABASE DEVELOPMENT AND MECHANISTIC STUDY

OF TRADITIONAL CHINESE MEDICINE BY COMPUTER

WANG JIFENG

NATIONAL UNIVERSITY OF SINGAPORE

2003
 Founded 1905

DATABASE DEVELOPMENT AND MECHANISTIC STUDY

OF TRADITIONAL CHINESE MEDICINE BY COMPUTER

BY

WANG JIFENG

A THESIS SUBMITTED FOR THE DEGREE OF

MASTER OF SCIENCE

NATIONAL UNIVERSITY OF SINGAPORE

2003
 Acknowledgment

First and foremost, I would like to express my sincerest appreciation to my supervisor,

Associate Professor Chen Yu Zong from Computational Science for his patient

guidance, supervision, invaluable advice and suggestions throughout the whole

research process.

Sincere gratitude is also expressed to Dr. Cai, Dr. Li, Xue Ying for their patient

guidance and cooperation. To Zhiwei, Zhiliang, Chenxin, Lizhi, Chunwei, Lianyi,

Chanjuan and Lixia, who are labmates as well as friends, for being ever so willing to

share with me their valuable advice on projects, and for sharing with my joy and

sorrow at all times.

I would like to thank Ms.Lucee, Ms Lindah, Ms. Hwee sim, Ms Elaine and Ms.Wei

Har,for their assistance and friendship.

Most of all, I am eternally grateful to my parents, my GF, for supporting me, and for

encouraging me at all times.

Finally, I would like to thank everyone in my department who had helped me in my

study.

Wang Jifeng

May 2003
Table of Contents

Content Page

List of Tables I

List of Figures II

Summary III

Chapter 1: Introduction 1

1.1 Brief History of Traditional Chinese Medicine (TCM) 1

1.2 Chinese Medicinal Herbs in TCM 4

1.2.1 Properties and Flavors 5

1.2.2 Meridians of Herb 9

1.2.3 Toxicity and Nontoxocity 11

1.3 TCM Formulae 12

1.3.1 Compatibility of Herbs 12

1.3.2 Precautions and Contraindications 15

1.4 Methods for Studying TCM 17

1.4.1 Theory and Practices of TCM 17

1.4.2 Modern Experimental Approach and Clinical Trials for

Studying TCM 18

1.4.3 Computational Methods 20

1.5 Specific Aims of the Project 21

1.5.1 To Develop a TCM Database 21

1.5.2 To Develop a Computer-aided Method

for Prescription Formulation 22

1.5.3 To Explore the Molecular Mechanism of Medicinal Herb 23

Chapter 2: TCM Database Development 24

2.1 Introduction 24

2.2 Database Development Method 25

2.3 Database Structure and Access 26

2.3.1 Database and Source of Data 26

2.3.2 Database Access 27

2.4 Data Submission and Update 32

2.5 Preliminary Analysis of Data 32

2.6 Conclusion and Future Development 34

Chapter 3: Development of a Computer-aided Method for

Prescription Formulation 36

3.1 Introduction 36

3.1.1 The Principle of TCM Prescription Formulation 36

3.1.2 Modification of TCM Prescription 38

3.1.3 Previous study on Prescription Formulation 40

3.2 A New Computer-aided Method for Prescription Formulation 41

3.2.1 Support Vector Machine (SVM) 42

3.2.2 Linear Classification 43

3.2.3 Nonlinear Classification 47

3.3 Dataset preparation 50

3.4 Feature vectors 50

3.5 Accuracy measure 56

3.6 Results and Discussion 57

Chapter 4: Exploration of Molecular Mechanism of a Medicinal

Herb Serenoa repens by IVDOCK 71

4.1 Introduction 71

4.2 INVDOCK Method 74

4.2.1 Protein Cavity Database 74

4.2.2 Inverse-docking Procedure 76

4.2.3 Scoring 78

4.2.4 Selection of Compounds and Therapeutic and Toxicity

Proteins 79

4.3 Results 84

4.3.1 Anti- inflammatory Effects 85

4.3.2 Anti-proliferate Effects 87

4.3.3 Anti-androgenic and Anti-estrogenic Effects 88

4.3.4 Arrest of Cell Cycle 91

4.3.5 Anti- metastasis 92

4.4 Discussion 92

4.5 Conclusion 98

Chapter 5: Conclusions 99

References 101
List of Tables and Figures

Tables Page

1. Properties and the Associated Effects of Herb 7

2. Flavors and the Associated Effects of Herb 8

3. Number of Positive Formulae and Negative Formulae in

Each Group 51

4. Principle for Constructing the Feature Vector 52

5. Example: Feature Vector of Herba Ephedrae (Ma Huang) 54

6. List of Positive Formulae in the Training and Testing Set

of Group 1
58

7. List of Positive Formulae in the Training and Testing Set

of Group 2
59

8. List of Positive Formulae in the Training and Testing Set

of Group 3
60

9. List of Positive Formulae in the Training and Testing Set

of Group 4
61

10. List of Positive Formulae in the Training and Testing Set

of Group 5
62

11. List of Positive Formulae in the Training and Testing Set

of Group 6
63

12. List of Positive Formulae in the Training and Testing Set

of Group 7
64

13. Number of Samples in the Training and Testing sets after 65

Calculation Using SVM

I
14. Sensitivity, Specificity and Overall Accuracy 66

15. False Predicted Negative Formulae (or Potential

Formulae) 69

16. Herbal ingredients of Serenoa repens 80

17. Predicted Proteins related with BPH 86

18. Other predicted important proteins 90

19. Summary of Compounds and the ir predicted targets 94

Figures Pages

1. The query interface of TCMID 28

2. The typical query result about formula 29

3. The typical query result about herb 30

4. The typical query result about compound 31

5. The data submission interface 32

6. Two possible separating hyperplanes 43

7. Definition of Hyperplane and Margin. 44

8. Schematic of the available Hyperplanes 45

9. Schematic of unique Optimal Separation Hyperplane 45

10. Illustration of basic principle of support vector machines 49

11. 3D Structure of Phytosterols of Serenoa repens 81

12. 3D Structure of Monoacylglycerides of Serenoa repens 81

13. 3D Structure of Fatty acids of Serenoa repens 82

14. 3D Structure of Ethyl Esters of Fatty acids

of Serenoa repens 83

II
Summary

Traditional Chinese medicine (TCM) has been used in the treatment of a variety of

diseases and is recognized as a valuable alternative to conventional medicine.

Increasing effort is being made towards scientific proof, clinical evaluation and

molecular study of TCM. To facilitate such an effort, I develop a database which

contains the available information about all major aspects of TCM, including herbal

formulations, herbal composition, chemical composition, molecular structure and

functional properties, therapeutic and toxicity effects, clinical indication and

application.

With the rapid development of computer technologies, computational methods

have been widely employed in biology. Support Vector Machine (SVM), based on

statistical learning theory, is such a method that has been used in a wide range of real-

world problems such as text categorization, cancer diagnosis, glaucoma diagnosis, and

microarray gene expression data analysis. In this study, SVM is used to facilitate the

study of TCM formulae. The results indicate the capability of SVM in recognizing

non-effective formulae and it may provide some helpful hints for herbalist doctors to

determine the effectiveness of a TCM formula. In addition, the computation provides

several potentially effective formulae from the hundreds of randomly mixed formulae.

It is unclear whether these formulae have the therapeutic value. The method is

expected to facilitate the prescription of new and novel TCM formulae as well as the

III
validation of existing TCM formulae while more and more formulae are under

scientific studies.

The mechanism of action of TCM remains largely unknown, though a large

number of active compounds have been isolated from these herbs and their clinical and

therapeutic effects have been probed. INVDOCK, a molecular interaction-based

method, is employed to study the molecular mechanism of medicinal herbs. This study

provides the potential targets of a medicinal herb Serenoa repens in the treatment of

BPH, parts of which have been demonstrated by previous experiments to be bound by

compounds in the extract. Besides these interactions, other bindings between particular

compounds and protein targets have not been proven by experiments. It provides a new

method for exploration of the mechanism of herb medicine. It is also of importance in

drug development based on herbs. In conclusion, as a relatively fast-speed and low-

cost tool, this method may find application in systematic study of the molecular

mechanism of multiple ingredients of other medicinal plants and has to be further

validated by clinical trials.

IV
 Chapter 1

Chapter 1: Introduction

1.1 Brief History of Traditional Chinese Medicine (TCM)

Traditional Chinese Medicine (TCM) has been used for thousands of years [1-5].

At least it has a recorded history dating back over 2,000 years. Among numerous

legends about the origins of traditional Chinese medicine, stories about three legendary

emperors/mythical rulers: Fu Xi, Shen Nong and Huang Di have to be mentioned.

Historians believe that Shen Nong and Fu Xi were early tribal leaders. Fu Xi is

regarded as a cultural hero who developed the trigrams of Yi Jing. Ancient books said

‘Fu Xi drew the eight trigrams, and created nine needles’. Shen Nong, the legendary

emperor who lived 5000 years ago, is regarded as the ‘Divine Farmer’ by the Chinese

people because of his attribution as the founder of herbal medicine. To test and analyze

the individual effects of different plant medicine, Shen Nong ingested them by himself.

It was said that Shen Nong tested over one hundred herbs including 70 toxic

substances in one day in order to find some drug to get rid of people's pain form

illness.

Huang Di Nei Jing (Yellow Emperor's Cannon of Internal Medicine) is the first

written documentation on traditional Chinese medicine, which was written between

800 BC and 200 BC [6]. This book summarized and systematized the previous

1
 Chapter 1

experience of treatment and theories of medicine, such as the meridian theory, as well

as many other issues, including physiology, pathology, prevention, diagnosis, treatment

and acupuncture etc. It was regarded as the foundation for the theories of Chinese

medicine.

During the Zhou dynasty, the theory of TCM was developing very fast. The most

important discoveries of medicine were made, including the theoretical foundations of

Yin and Yang [7], the five elements, the pathogenic factors of external environment as a

cause of disease and further understanding of the meridians of acupuncture.

The basic theories of acupuncture were established and stone needles became

obsolete, being replaced by metal needles. At the time of the Spring and Autumn

Warring States Period, one of the most important issues in the development of TCM

was the usage of the pulse for diagnosis. Bian Que [8], a very famous doctor/physician

was the first man in the world to use this technique. He was reputed to be an excellent

diagnostician, excelled in using acupuncture and moxibustion, boiled herbal

prescriptions, and massage in internal medicine, external medicine, gynaecology, and

paediatrics for the treatment of all kinds of illnesses. Bian Que also recorded his

experiences in the book Nan Jing (The Classic of Difficult Issues), which developed

and explained the fundamental and difficult parts of Huang Di Nei Jing.

During this Period another famous classic book Shen Nong Ben Cao Jing (Shen

Nong's Classic of Materia Medica) [9] was written. This book recorded and described

the characters of about 365 herbs and was regarded as the first pharmacopoeia of

2
 Chapter 1

traditional Chinese herbal medicine.

The Han dynasty, which lasted almost four and a half centuries, was a period of a

thriving Chinese culture. TCM was well developed as well. Zhang Zhongjing (150-219

A.D.), one of the most famous herbalist doctors, was reputed for his remarkable

medical skill and his well known medical masterpiece Shang Han Lun (Treatise on

Febrile Diseases) [10], which was used as a standard reference work for traditional

Chinese medicine, including moxibustion, needling and herbal medicine. So far his

theory and prescriptions are still of great practical value. In this period there also was

one famous physicians/surgeon of traditional Chinese medicine, Hua Tuo (110-207

A.D). He invented the use of anaesthesia called Ma Fei San [11] to reduce the pain of

patients who was suffering surgery. He also furthered the knowledge of anatomy. He

was the first person who used narcotic drug in the world and his skill in this field was

ahead of the west about 1600-1700 years.

Li Shizhen was considered to be China's greatest scholar in TCM and made great

contribution to the progress of TCM. He spent forty years to collect and taste herbal

medicines and wrote down in his most well-known Chinese herbal book Ben Cao

Gang Mu (Herbal Systematics) [12,13], which contained 1892 different herbs (with

1110 drawings), and was divided into 6 sections, 52 scrolls and 60 different categories

and also included 11,096 prescriptions, for treating hundreds of illnesses, ranging from

the common cold to drunkenness and food poisoning. Others such as Wang Shuhe,

Huang Fumi, Ge Hong, Chao Yuanfang, Sun Simiao and Wa ng Weiyi also wrote

important medical manuals and contributed to the thriving of TCM.

3
 Chapter 1

Indeed, TCM has a long history, and in what is regarded as a breakthrough, the

World Health Organization (WHO) in 1979 released a list of 43 types of pathologies,

which can be effectively treated with acupuncture. Today, there are many hospitals in

China, Japan, and the other Asia countries that practice TCM exclusively, and others

that combine Eastern and Western healing methods. One of the main reasons that TCM

is still used may be their relatively low side effects, compared with western medicine.

Another reason may be that it is used as a last resort, when Western medicine is too

toxic or unable to provide the expected benefit [14]. In recent years, the effectiveness

of TCM has been gaining popularity in US and European countries, such as Germany,

France and UK [2,15]. However, there are also negative attitude toward TCM. Some

westerners think that TCM is unscientific in its understanding of the human body and

the nature of disease and its treatment. They believe that the lack of quality control and

the absence of scientific and clinical proof of their effectiveness will impede the

adoption of TCMs in industrialized countries [1,16-18].

1.2 Chinese Medicinal Herbs in TCM

As the most important parts of TCM, medicinal herbs and the prescriptions of

multiple herbs, which will be discussed in the next chapter, were given much attention

in the development of TCM [4,5]. In TCM, people think that the actions of herbs in

treatment of diseases is through rectifying the balance of Yin or Yang, and then

consequently helping the body restore its normal physiological functions. Different

herbs have different characters and functions. That is why different herbs are used

when curing different diseases. To understand the mechanism of them and use them

4
 Chapter 1

properly, it is necessary to study and explain their characters. Ancient experts had,

from the TCM viewpoint, summarized those characters, which include drugs'

Properties and Flavors, Meridians of herbs and Toxicity property, etc. Based on the

theories of Yin and Yang, Viscera, Channels and Collaterals, and treatment principles of

traditional Chinese medicine, the information has been summed up throughout a long

history of medical practice.

1.2.1 Properties and Flavors

Each herb has its own properties and flavors. Generally, there are four typical types

of properties, that is, cold, hot, warm or cool [19]. These properties are experientially

summarized according to the actions of the herbs on the human body. Herbs that cure

or reduce heat syndrome (Yang syndrome) have a cold or cool property, whereas herbs

that cure or reduce cold syndrome (Yin syndrome) have hot or warm property. Cold or

cool properties are quite different from warm or hot. People think that cold and cool

are similar, and so are warm and hot. The difference is merely the action abilities. Cold

is relatively ‘stronger’ than cool and hot is ‘stronger’ than warm. Cool- or cold

-natured herbs are thought to have the effects of clearing heat, purging fire, removing

toxic substances, and nourishing Yin, so they are usually used to cure heat syndromes.

On the contrary, warm and hot –natured herbs are believed to have the effects of

dispersing cold, warming up the interior, supporting Yang, and treating collapse, and

are therefore used to treat cold syndromes. In addition to the four properties mentioned

above, there is the fifth property: mild. A mild- natured herb can be used for either hot

or cold syndromes.

5
 Chapter 1

Flavor of herbs is given partly by their tastes. Sometimes they are given according

to the actions of herbs rather than tastes. Therefore, the flavors of some herbs are often

different from their true tastes. There are a total of seven flavors, including pungent,

sweet, sour, bitter, salty, tasteless and astringent [19]. The first five ones are the basic.

Herbs with different flavors usually show different pharmacological and therapeutic

actions, while the same flavor comes out the similar effects. According to Yin and Yang

theory, herbs with pungent, sweet or tasteless flavor have the attribution of Yang and

the ones with sour, bitter or salty flavor, of Yin. The effects of pungent herbs are to

disperse exopathogens from the body and promote the circulation of the vital energy

and blood. Therefore, these herbs, such as Herba Ephedrae (Ma Huang), Radix

Aucklandiae (Mu Xiang), are often used in treatment of superficial and mild illnesses.

The effects of sweet herbs are to nourish, replenish, or enrich the function of the

organs, to normalize the function of the stomach and spleen, to harmonize the

properties of different herbs and to relieve spasm and pain, etc. They are usually

effective in treating syndromes of deficiency type, dry cough, constipation, such as

Radix Codonopsis pilosulae (Dang Shen), Radix Rehmanniae Preparata (Shu Di) and

Radix Glycyrrhizae (Gan Cao). The effects of sour herbs are to induce astringency and

arrest discharge. They are often used to treat sweating. For example, Fructus

Schisandrae (Wu Wei Zi), a very important adaptogen can be used to regulate body

functions and increase the organism’s ability to deal with stress. Herbs with bitter

flavor have the effects of clearing heat, purging fire, sending down the adverse flow of

Qi to treat cough and vomiting, relaxing the bowels, eliminating dampness, etc. Such

6
 Chapter 1

herbs are mostly used for syndromes of pathogenic fire, cough with dyspnea, vomiting,

constipation due to heat of excess type, damp- heat syndrome, or cold-damp syndrome

and other syndromes. Herbs with salty flavor have the effects of relieving constipation

by purgation, and softening and resolving hard mass. They are used in the treatment of

dry stool and constipation, scrofula, goiter and mass in the abdomen. Two examples

are Concha Arcae (Wa Leng Zi) and Natrii Sulfasl (Mang Xiao). The effects of

tasteless herbs are to excrete dampness and induc e diuresis. Therefore these herbs,

such as Polyporus umbellatus (Zhu Ling) and Poria (Fu Ling), are often used for

edema and dysuria. Astringent herbs have similar actions as those sour herbs, such as

Os Draconis (Long Gu), Concha Ostreae (Mu Li) and Halloysitum Rubrum (Chi Shi

Zhi).

Table 1. Properties and Associated Effects of Herb

Property Effects

Cold Quells fire (anti- inflammatory/spasmodic, sedative).

Cool Subdues heat (reduces fever, detoxifies, lowers BP).

Mild Gentle effects (does not alter Hot or Cold conditions).

Warm Enhances circulation (alleviate chills, improves organ function).

Hot Dispels Cold (breaks Qi blockage, warms the center).

Note: Cool and cold herbs do overlap, as do hot and warm herbs.

7
 Chapter 1

Herbs that possess the same flavors and properties generally have similar effects.

But if only one property or flavor is the same, their actions may be quite different. For

example, both Rhizoma Coptidis (Huang Lian) and Radix Rehmanniae Preparata

(Sheng Di Huang) are cold, however the former is bitter and the latter is sweet. It has

been well known about the effects of these two herbs. Rhizoma Coptidis has the effects

of clearing heat and drying dampness and is therefore used for damp-heat syndrome,

while Radix Rehmanniae Preparata has the effects of clearing heat and nourishing Yin

and is used for the condition of consumption of Yin due to febrile diseases. Table 1 and

Table 2 give the four typical properties and five typical flavors, respectively, and the

corresponding effects.

Table 2. Flavors and Associated Effects of herb

Flavor Effect

Sweet Nourishing, tonifying.

Pungent Dispersing, decongesting, stimulating.

Salty Diuretic, purgative, softening

Sour Astringent, absorbing, circulation.

Bitter Sedating, anti- inflammatory/Fire, soothing.

8
 Chapter 1

1.2.2 Meridians of Herb

According to Viscera, Channels and Collaterals theory, the symptoms can reflect

the organs that are not in good conditions. So ancient herbalist doctors speculated that

a herb might selectively act upon a particular part of the body and this part depended

on the corresponding symptoms that the herb can relieve. These parts of the body are

named meridians. There are twelve types of merid ians [19], including Lung, Bladder,

Spleen, Large Intestine, Stomach, Small Intestine, Liver, Pericardium, Heart, Kidney,

Gallbladder and San Jiao. The Lung is the most delicate and most exterior of all the

organs. Diseases often happened because of external pathogenic invasion and

accumulation of Lung heat. The Bladder transforms and excretes fluids from the body. It

is extremely sensitive to climatic changes, which can cause induced patterns of

Cold-Damp and Damp-Heat. The Spleen is the primary source of nourishment of the

body because it governs digestion and the production of Qi. It is most susceptible to the

evil of Dampness. The Large Intestine is used to receive food from the Small Intestine,

absorb fluids and excrete feces. When excessively exposed to cold, External Cold can

invade it resulting in abdominal pain and diarrhea. The Stomach has the function to

digest and transform food to make it available to the Spleen. The state of the Stomach is

governed by the Hot-Cold nature of ingested food in rela tionship to the patient’s

constitution and the environment. The Small Intestine receives the transformation

products of food and drink from the Stomach and separate the pure from the impure. It is

also affected by the heat or coldness of ingested food and drink. The liver is the central

organ of the body, which is primarily responsible for the storage of Blood and the

9
 Chapter 1

smooth distribution of Qi throughout the body. Exterior Wind can interfere with the

smooth flow of Qi and stir up the Blood stored in the Liver to exacerbate Internal Liver

Wind and cause skin rashes of sudden onset that move around the body, often seen in

viral exanthema, drug rashes and hives. The Pericardium is membranous sac filled with

serous fluid that encloses the heart and the roots of the aorta and other large blood

vessels. Disease can be caused by cold, heat and Liver Qi stagnation. The heart controls

the blood vessels and regulates the flow of blood in the body. The kidney represents

the most important energetic organ in the body with regard to the root of energy and

the will to live, develop and reproduce. It is the foundation of Yin and Yang in the body.

Fear, anxiety, shock and prolonged stress are the major emotional issues that deplete

Kidney energy. The Gallbladder is used to store and secrete bile. The Excessive

ingestion of greasy or fatty foods leads to Dampness that becomes lodged in the

Gallbladder. The emotions of anger, repressed resentment and frustration cause Heat

and Liver and Gallbladder Fire, and when mixed with Dampness, give Damp-heat.

Climatic changes of Heat and Dampness from the exterior will induce elements of

Damp-Heat in the Gallbladder. San Jiao refers to Shang Jiao, Zhong Jiao and Xia Jiao,

which are the pathway through which Qi and Jing Ye ascend or descend. Disease can

be caused by Wind-Damp, Cold-Damp and Heat.

By acting on one or more particular meridians, medicinal herbs can regulate the

body to a balance status. For example, Fructus Zizyphi Jujubae (Da Zao) can tonify Qi

in the spleen and stomach. It is indicated for poor appetite and loose stool due to

weakness of the spleen and stomach. Judged by the above indications and analysis, we

10
 Chapter 1

say that the herb enters the meridians of the spleen and stomach. Herba Ephedrae (Ma

Huang) can promote sweating, soothe asthma and benefit urination. It is used for fever,

chills and absence of sweating due to invasion by exogenous pathogenic wind and cold,

dysuria, edema and so on. So people believe that the herbs can enter the lung and

bladder meridians.

The meridian theory studies the physiological function and pathological change on

the meridians and the related zangfu organs. The essential functions of the meridian

system are to transport Qi and blood, to maintain conductivity and to resist invasion of

exogenous patho genic factors. The meridian system distributes to all parts of the body.

The endless circulation of Qi and blood in the meridians is responsible for the

maintenance of life and the variety of functions which support it. The meridian theory

has been the guiding principle for the clinical practice in the realms of TCM,

particularly in those of acupuncture, massage and Qi Gong. By combining the

meridian theory with the theories of the zangfu organs and the etiology of TCM, one

can thoroughly explain both the physiological activities and pathological change s,

which take place in the body. In this manner, a theoretical basis for the principle of

treatment in accordance with the differentiation of symptoms and signs was

established.

1.2.3 Toxicity and Nontoxicity

Traditionally, people, especially those in Chinese communities, believed that

medicinal herbs were weakly toxic or even non-toxic. However, some herbs are toxic

11
 Chapter 1

and should also be used carefully. According the difference of their toxicities, they are

classified to three classes: toxic, extremely toxic and slightly toxic [19]. Improper use

of the toxic herbs may lead to adverse effects, so prescription of toxic herbs should be

careful according to the patient’s age and the situation of the disease, etc. Nontoxic

herbs are moderate in nature and usually do not have any side effects. For example,

Fructus Zizyphi Jujubae (Da Zao) and Poria (Fu Ling) are nontoxic herbs, while Radix

Aconiti lateralis (Fu Zi) and Semen Strychni (Ma Qian Zi) are toxic.

Some toxic herbs are effective because of the other effects on the patient, so

when used in combination of other herbs in a prescription, the toxicities are expected

to be eliminated or lessened by means of processing, dispensing and preparation. On

the contrary, the medical effects of some toxic herbs, on the critical or obstinate

diseases, are due directly to their toxicity properties.

1.3. TCM Formulae

1.3.1 Compatibility of Herbs

Two or more herbs are often combined in order to increase or promote their

therapeutic effectiveness, to minimize toxicities or side effects, to accommodate

complex clinical situations and to alter their actions. Different combinations can cause

diverse therapeutic effects. Using only one herb in a formula is called “Single herb

formula”. The combination of two or more herbs is known as “mutual reinforcement,

mutual assistance, mutual restraint, mutual counteraction, mutual suppression and

12
 Chapter 1

mutual antagonism” [20]. The details for combining herbs are classified as follows.

1. Single herb formula. The whole formula is composed of only one herb. This

formula is used when the disease is not very serious. For example, Formula ‘Du Yi Wei

Wan’, which only contains Radix Lamiophlomidis rotatae (Du Yi Wei), is used in

treatment of pain caused by traumatic injury, sprain, rheumatism, contusion of muscles,

joints and loins, fracture, surgical injury and rheumarthritis.

2. Mutual reinforcement. Two or more herbs with similar properties are combined

to reinforce their therapeutic actions. For example, Radix et Rhizoma Rhei (Da Huang)

is combined with Natrii Sulfasl (Mang Xiao) to reinforce the function of purging

downward; Gypsum Fibrosum (Shi Gao) and Rhizoma Anemarrhenae (Zhi Mu) are

used together to clear heat and subdue fire.

3. Mutual assistance. Two or more herbs in which one or more is the principle herb

and the others play a subsidiary role are combined. For example, Radix Astragali

(Huang Qi) is combined with Poria (Fu Ling) to replenishe Qi, strengthen the spleen

and promote urination; Gypsum Fibrosum (Shi Gao) is combined with Radix

Achyranthis bidentatae (Niu Xi) to relieve toothache.

4. Mutual restraint. In the formula, the toxicity or side effects of one herb can be

reduced or eliminated by the addition of another. For example, the toxicity of Rhizoma

Pinelliae (Ban Xia) can be counteracted or restrained by Rhizoma Zingiberis recens

(Sheng Jiang).

5. Mutual counteraction. It means that one herb in the formula can counteract or

13
 Chapter 1

restrain the other herb’s toxicities and side effects. In fact, mutual restraint and mutual

counteraction refer to the same effect of herb from two different aspects. The former is

focused on how to restrain the toxicity of one given herb, while the latter pays

attention to the counteraction effects of one herb on the other.

6. Mutual suppression. In the formula, one of the herbs weakens or suppresses the

action of the others. For example, Semen Raphani (Lai Fu Zi), combined with Radix

Ginseng (Ren Shen), weakens the function of the latter in replenishing Qi.

7. Mutual antagonism. When two herbs that have no side effects are combined

together, severe side effects may result. Traditionally, there existed “eighteen

incompatible medicinal herbs” and “nineteen mutual restraining medicinal herbs”.

When two herbs are combined together in one formula, they may give rise to

interaction with each other. They may be mutual reinforcement, mutual assistance,

mutual restraint, mutual counteraction, mutual suppression and mutual antagonism.

For this reason, combination should be carefully considered according to the

conditions of the patient, and their characters and functions. “Mutual reinforcement”

and “mutual assistance” can enhance their effects and therefore should be used as

much as possible. “Mutual restraint” and “mutual count eraction” can reduce or

eliminate toxicities and side effects of herbs and therefore can be considered when

using poisonous herbs. “Mutual suppression” and “mutual antagonism” can weaken

efficacy of herbs or make them lose their efficacy or even give rise to toxicities and

side effects and therefore should be avoided.

14
 Chapter 1

1.3.2 Precautions and Contraindications

The side effects of TCM formula can be reduced by adding one or two herbal

ingredients. But sometimes it would be harmful to the patients [20]. So the prescription

should be carefully given out according to condition of the patients. Four

contraindications should be noticed as following:

1. Contra-syndromes. Each herb or each class of herbs has its own intended

functions and effects. Diseases or syndromes other than these intended indications are

contra-syndromes or contraindications. For example, Herba Ephedrae (Ma Huang) is

used to induce diaphoresis and relieve asthma, and its indications are affection by

exopathogenic wind-cold, anhidrosis due to exterior syndrome of excess type and

cough due to obstruction of the lung Qi, but in case of spontaneous sweating due to

exterior syndrome of deficiency type or cough due to lung deficiency, its use should be

prohibited.

2. Incompatibility of herbs. Some herbs cannot be used in combination with

specific herbs. According to the ancient books and literature of TCM, there are

eighteen pairs of incompatible medicinal herbs and nineteen pairs of mutual restraining

medicinal herbs.

3. Contraindication for pregnancy. Some herbs cannot be used by pregnant patients.

These herbs are mainly with strong actions or toxicities, especially Fructus Crotonis

(Ba Dou), Semen Pharbitidis (Qian Niu Zi), Herba Cirsii japonici (Da Ji) and Rhizoma

Sparganii (San Leng). The other herbs that are pungent and hot should also be used

15
 Chapter 1

with caution during pregnancy, such as Semen Persicae (Tao Ren), Flos Carthami

(Hong Hua), Radix et Rhizoma Rhei (Da Huang) and Radix Aconiti lateralis (Fu Zi).

4. Food taboo. Certain types of food may also influence the action of herbs or

induce unnecessary side effects. In general, it is advisable not to eat raw, cold, greasy,

strong smelling or spicy food while taking medicine. From the historical medical

literature, Radix Dichroae (Chang Shan) is contraindicated with onion; Radix

Rehmanniae (Di Huang) and Radix Polygoni multiflori (He Shou Wu) are

contraindicated with onion, garlic and turnip; Herba Menthae (Bo He) is

contraindicated with turtle meat; Poria (Fu Ling) is contraindicated with vinegar; and

Carapax Trionycis (Bie Jia) is contraindicated with three-colored amaranth.

Each herb has its own functions and indications. The herbs should be selected

according to their properties, flavors, meridians and toxicity properties. For example,

Herba Ephedrae (Ma Huang) is pungent, slightly bitter and warm. It is used to treat

fever, chills and absence of sweating due to an invasion of exogenous wind and cold. It

is contraindicated in deficiency exterior syndrome with the symptoms mentioned

above.

In one formula, the different dosage of each herbal component is prescribed

according to their effects [20]. Generally speaking, the dosage of the principal

component is bigger than the others. When this formula is used for the treatment of

patients with different conditio ns, the dosage of these herbs may be changed in

accordance with specific syndrome. In addition, the dosage of those herbs that are very

16
 Chapter 1

drastic in nature or extremely poisonous should be strictly controlled to prevent

poisoning.

1.4 Methods for Studying TCM

1.4.1 Theory and Practices of TCM

TCM practice is believed to be based on the principle of maintaining the balance of

the body. Illness or disease always represents a state of imbalance of Yin and Yang. We

could consider this concept as the need to find the most appropriate and stable chaotic

state, with a strong attractor. Over time, TCM has elucidated methods of restoring

balance to the body, thus curing disease. The two important methods are herbal therapy

and acupuncture. Although various kinds of imbalances can be treated by either

method, herbal medicine is excellent at tonification, that is, increasing the energy of a

particular organ, and balancing the energy when there is excess in one part of the body

and deficiency in another. Of course, in integrated medicine we can also compile

acupuncture and herbal medicines together – acupuncture for the acute problem, and

herbs to shore up the underlying deficiency that allowed the acute disease to manifest.

Ancient Chinese herbalist doctors have developed their own language and method

for diagnosis. The diagnosis is not directly associated with a disease, but the

description of a pattern that seems to be preeminent in the whole body, which includes

Looking, Smelling, Asking, Pulse diagnosis, etc [21]. The Chinese physical exam is

able to show a predominant pattern of imbalance in the patient, which, in the Chinese

physician’s eyes, is the disease to be treated. Many dissimilar western diagnoses have

17
 Chapter 1

similar Chinese patterns of disharmony, and many similar Western diagnoses have

different Chinese patterns of disharmony. In Chinese medicine there have been periods

in which one theory of harmony and disharmony dominated over another. Thus, during

one period, the Five Element Theory of disharmony reigned supreme and during

another period, the Yin- Yang and Eight Principles Theory of disharmony reigned

supreme. In this paradigm of healing, the western diagnosis is unimportant, as we are

treating the patient ’s explicit response to his or her inner disturbance, which is going to

be different for each patient according to his or her initial condition.

1.4.2 Modern Experimental Approach and Clinical Trials for Studying TCM

From the point of view of modern medicine, the effects of herbal drugs result from

the interaction between herbal ingredients and organs of body [2,5]. It is necessary to

extract these chemicals and then analyze their physiological effects on humans or their

action on disease. So far, a considerable body of information on the chemical

composition of medicinal herbs, has been found by researchers and institutes,

especially those in China and Japan. Many purified compounds are then studied or

subjected to be studied in vitro or on animal models [22]. For example, Radix

Glycyrrhizae (Gan Cao) is one of the most extensively studies Chinese herbs [23-28].

The function of Radix Glycyrrhizae is to reinforce the function of the spleen and

replenish Qi, to remove heat and counteract toxicity, to dispel phlegm and relieve

cough, to alleviate spasmodic pain, and to moderate drug actions. It is used in

treatment of weakness of the spleen and the stomach marked by lassitude and

weakness, cardiac palpitation and shortness of breath, cough with much phlegm,

18
 Chapter 1

spasmodic pain in the epigastrium, abdomen and limbs, carbuncles and sores. It is

often used for reducing the toxic or drastic actions of other drugs. Its main ingredients

are Glycyrrhizin and glycyrrhetinic acid [25,26]. Experiments show that its extract

exerts a detoxifying effect [26,27]. Another experiment shows that glycyrrhizin also

has this effect [26,27]. In addition, Glycyrrhizin has anti- inflammatory effect similar to

hydrocortisone [28]. It was found that its decoction can inhibit intestinal smooth

muscles of rabbits in vitro, but glycyrrhizin and glycyrrhetinic acid have no such

effects [29]. Thus the effects must be due to the other compounds. Radix Ginseng (Ren

Shen) being under exploration [30,31], contains more than 200 compounds. The main

ingredients are ginsenosides, essential oils and polyacetylenes [30]. Radix Ginseng

traditionally has been used as a tonic and was thought to reinforce the vital energy, to

remedy collapse and restore the normal pulse, to benefit the spleen and lung, to

promote the production of body fluid, and to calm the nerves. Experiments have shown

that Radix Ginseng can act on the central nervous system and exert significant

cardiotonic and hypertensive effects on acute circulatory failure cardiotonic and

hypertensive effects on acute circulatory failure after heavy blood loss and decrease

the level of blood sugar and promote phagocytosis and enhancing

lymphocyte-blastogenesis rate [29]. These effects are related to the active compounds.

For example, ginsenoside Rg1 can raise blood pressure, while Rb1 can lower it

[31,32].

As indicated in the above examples, chemical and biological studies underlined the

importance of using both together in the standardization and modernization of TCM.

19
 Chapter 1

The chemical profiles of an extract should be matched against the results of biological

assays in order to assure lot-to-lot consistency of what are fairly crude extracts of herbs

with high natural variability [1]. Chemical analysis, biological assays and animal

experiments provide important fundamental information about TCM. In the context of

modern biomedical research, they should also be necessary prerequisites for clinical

trials. Most of these TCM clinical trials have been undertaken in China, Taiwan, and

Japan. A number of commercial treatments based on TCM such as ‘Hu Xin Dan”,

“Puerarin injection’, ‘Shen Mai injection’ and others have been introduced into the

market after extensive pharmacological research and clinical trials [33,34].

1.4.3 Computational Methods

Conventional approaches, including chemical standardization, biological assays,

animal models and clinical trials, have been extensively used to study the effects and

molecular mechanism of medicinal herbs and formulae. More recently, as useful tools,

computer-aided methods are rapidly developed to facilitate the understanding of TCM.

To control and evaluate the quality of medicinal herbs, the fingerprint analysis

technique is set up with the rapid development of instrument analyses and computer

pattern interpretation, which includes fuzzy information analysis, artificial neural

networks and gray relational grade cluster [35-37]. To identify similar medicinal herbs,

computer-aided classification based on back propagation algorithm of artificial neural

network pattern recognition is developed [38].

Employing a fuzzy clustering method, Su Weiwei et al divided the herbs in one

formula into several different parts [39-41]. The results agreed well with the

20
 Chapter 1

explanation by TCM theories. Moreover, it provided a feasible computer-aided method

to study the formulation of prescription. Aiming at this problem, in this study, we

proposed a new method based on Support Vector Machine (SVM) as discussed in

Chapter 3 to evaluate the effectiveness of a formula and to generate new potential

formulae.

To understand the molecular mechanism and pharmacology of bioactive

compounds from Chinese medicinal herbs, which is important in facilitating scientific

evaluation of novel therapeutic approaches in traditional Chinese medicine, a newly

developed computer software INVDOCK [42] is used for automated identification of

potential therapeutic and toxicity targets of bioactive compounds isolated from

Chinese medicinal herbs. This method may potentially be used as a relatively

fast-speed and low-cost tool for facilitating the study of molecular mechanism and

pharmacology of bioactive compounds from Chinese medicinal herbs. And it is also of

significance in new drug development based on the mechanism of Chinese medicine.

1.5 Specific Aims of the Project

1.5.1 To Develop a TCM Database

Due to the difficulties in accessing and understanding non- English medicinal

records, most scientists are not familiar with traditional Chinese medicine [1].

Moreover, because the TCM theory is very different from that of western medicine,

majority of these scientists show little interest in TCM and other traditional medicines.

21
 Chapter 1

To attract their attention and help them better understand the merits of TCM, it is

necessary to provide the information of TCM first. Hence, a new database, Traditional

Chinese Medicine Information Database (TCMID), is developed, to provide, in an

integrated manner, comprehensive information about all aspects of TCM, including

prescription formulae, constituent herbs, known herbal ingredients and their molecular

actions. The therapeutic effects, adverse reactions, clinical indications and applications

are provided at the levels of prescription, individual herb and herbal ingredient

respectively. Traditional terminologies about medicinal herbs such as properties,

flavors, sites of actions, toxicity level, and therapeutic class are also given.

1.5.2 To Develop a Computer-aided Method for Prescription Formulation

In recent years, a great deal of pharmacological research has been undertaken to

review and establish reliable composite formulae of TCM [1]. To facilitate such a study,

an alternative approach is introduced, based on a computer method, support vector

machines (SVM), which is a machine learning algorithm currently considered to be

one of the most efficient method in many real-world applications [43]. Recently SVM

have been applied to a number of biological problems [44], including gene expression

data analysis and protein classification, etc. Using this method, it is expected to find

the principle of prescription formulation and then to evaluate the effectiveness of a

given formula and even generate new formulae.

1.5.3 To Explore the Molecular Mechanism of Medicinal Herb

The mechanism of action of TCM remains largely unknown, though plenty of

22
 Chapter 1

active compounds have been isolated from these herbs and clinical and therapeutic

effects are found. This study is proposed to explore the mechanism by a

computer-aided method, INVDOCK. In this study, I analyzed the mechanism of

Serenoa repens, which is widely used in treatment of BPH [45-49]. Although the effect

of some active compounds extracted from it has been proven by in vivo or in vitro

experiments, the whole picture is still unknown. The multiple ingredients make the

process more difficult. Hopefully, the effects of the predicted therapeutic and toxicity

targets will provide us clues of the molecular mechanism of this herb.

23
 Chapter 2

Chapter 2: TCM Database Development

2.1 Introduction

Traditional Chinese medicine (TCM) has been used in the treatment of a variety of

diseases and is recognized as a valuable alternative to conventional medicine

[1,2,16,50]. A major therapeutic approach of TCM is the use of a mixture of herbs,

each composed of a number of constituent chemicals, which collectively exert

therapeutic actions and modulation of other factors. One or more principal constituents

provide main therapeutic actions. Certain secondary principal constituents enhance or

assist the effect of the principle ones. The rest serve s modulation roles such as

treatment of accompanying symptoms, moderation of harshness and toxicity,

enhancement of drug delivery, and harmonization etc [1].

Because of growing interest in TCM therapeutics, increasing effort has been made

towards scientific proof, clinical evaluation and molecular study of TCM [1,2,42].

Such an effort can be facilitated by making available information about all major

aspects of TCM including herbal formulations, herbal composition, chemical

composition, molecular structure and functional properties, therapeutic and toxicity

effects, clinical indication and application. A few specialized databases have appeared

which provide information about different aspects of TCM. For instance, The TCMD

24
 Chapter 2

provides information about Chinese medicinal plants and constituent chemicals [51],

The Chinese Medicine Sampler (http://www.chinesemedicinesampler.com/) and

Traditional Chinese Medicine

(http://www.healthy.net/CLINIC/therapy/Chinmed/Index.asp) databases provide

general information about the history, theory, diagnostics, and examp les of TCM

formulae.

However, there is a lack of a database that provides comprehensive information

about all major aspects of TCM. Moreover, in the existing databases, information

about the composition of constituent chemical compounds in TCM herbs is not

included. A new database, Traditional Chinese Medicine Information Database TCMID,

is introduced to provide, in an integrated manner, comprehensive information about all

aspects of TCM including prescription formulae, constituent herbs, known herbal

ingredients and their molecular actions. The therapeutic effects, adverse reactions,

clinical indications and applications are provided at the levels of prescription,

individual herb and herbal ingredient respectively. Traditional terminologies about

medicinal herbs such as tastes, flavors, sites of actions, toxicity level, and therapeutic

class are also given.

2.2 Database Development Method

In this work, Oracle 9i is selected as the platform for our database projects and data

future analyzing tasks. Oracle 9i is a typical database management software, which has

the most variety of modules and development tools, including modules for data mining

25
 Chapter 2

and online analytical process, which is essential for high-end data analysis purpose.

For many years, it leads the way in indexing and query optimization technology while

it is not worse than its competitors in other aspects. Also, Oracle 9i is a fully

object-oriented database, which conforms to the world trend toward object, oriented

programming (OOP). Oracle has kept the biggest market share for years. According to

2002 statistics (http://www.oracle.com/tellmemore/?1333278), over half of the fortune

100 corporations use Oracle as their database server.

Oracle 9i is based on a relational approach, which means that database developer

can create a set of tables (also known as relations) to reflect the inter-relationship

between the data stored in the database. The relational approach needs the support of a

relational database management system (RDBMS). The RDBMS enables users to

create and maintain a relational database. They are designed to control data redundancy,

to restrict unauthorized access, to provide persistent storage for program objects and

data structures, to permit inference and actions using rules, to provide multiple user

interfaces, to representing complex relationships among data, to enforce integrity

constraints, and to provide backup and recovery.

2.3 Database Structure and Access

2.3.1 Database and Source of Data

TCMID is freely accessible at http://xin.cz3.nus.edu.sg/group/tcmid/tcmid.asp. The

information collected in TCMID is from a search of available literature [29,52-62].

Efforts are made to collect as many of these known formulae, herbs, herbal

26
 Chapter 2

constituents and compositions, and other functional information as possible. Intensive

research into TCM and TCM herbal constituents are leading to the discovery of new

knowledge about different aspects of TCM. It is hoped that new or existing

information missing in this database can be gathered by users’ submission to a special

page in the database as well as by conducting up-dated literature search.

The database currently contains entries for 1343 formulae, 1325 medicinal herbs,

4468 herbal ingredients and description about molecular actions of 502 herbal

ingredients. Each entry can be retrieved via the name, alternative name, function or

clinical manifestation at the levels of prescription formula, herb and herbal ingredient

respectively. This database will be continually updated as new data are found. And the

friendly interface will be upgraded as well to meet the user’s needs adequately.

2.3.2 Database Access

The TCMID database web interface is shown in Figure 1. Users can easily search

the database from three starting points, namely, Formula, Herb and Compound.

‘Formula’ starting point generates a new interface through which the database can be

accessed by keyword search such as Formula Chinese Names, Formula English Names,

Formula Functions, Formula Clinical Manifestations, etc. ‘Herb’ starting point

generates another interface to access the herbal information through keyword search

such as Herb Latin Names, Herb Chinese Names, Herbal Functions, Herbal Clinical

Manifestations, etc. ‘Compound’ starting point generates the third interface to access

the chemical information through keyword search such as Compound Name, CAS No,

27
 Chapter 2

etc. Searches involving any combination of these search or selection fields are also

supported.

Figure 1. The query interface of TCMID

The full text search is case insensitive and wildcards supported. In a query, a user

can specify full name or any part of the name in a text field. Wild character of ‘%’ and

‘_‘ is allowed in text field. Here, ‘_’ represents any one character and ‘%’ represents a

string of characters of any length. For example, input of ‘Ma Huang’ in the Formula

Chinese Name field finds entries containing ‘Ma Huang’ in their name, such as Ma

Huang Tang or She Gan Ma Huang Tang. On the other hand, input of ‘Ma% Tang’

finds all the entries with ‘Ma’ and ‘Tang’ in their names, such as Ma Huang Tang.

The typical query results are illustrated in Figure 2, 3 and 4. Figure 2 gives

details about the formula, including Chinese Name, Common Name, Literature

28
 Chapter 2

Reference, Herbal Components, Function, Clinical Manifestation, Procedure,

Description, Usage and Dosage, Storage, Precaution and Modifications. From this

page, the herbal information can be obtained by clicking each individual herbal

component.

TCMID Test Result (Formulae Info)

Chinese NameGe Jie Ding Chuan Wan
Common
Gejie Dingchuan Pills, Asthma-Relieving Bolus of Gecko
Name
Gecko: 11 g; Semen Trichosanthis: 50 g; Radix Asteris : 75 g;
Herba Ephedrae: 45 g; Carapax Trionycis : (processed with
Vinegar) 50 g; Radix Scutellariae: 50 g; Radix Glycyrrhizae:
Herbal
50 g; Radix Ophiopogonis: 50 g; Rhizoma Coptidis : 30 g;
Components
Bulbus Lilli: 75 g; Fructus Perillae: (stir-fried) 25 g; Gypsum
Fibrosum: 25 g; Semen Armeniacae Amarum: (stir-fried) 50 g;
Gypsum Fibrosum: (calcined) 25 g.
Pulverize the above fourteen ingredients to fine powder,
sift and mix well. To each 100 g of the powder add 70-100
Procedure
g of refined honey to make small honeyed pills or big
honeyed pills.
Blackish-brown small honeyed pills or big honeyed pills.
Description
Odour,slight;taste,bitter and sweet.
Function To nourish yin and cool lung, relieving cough and asthma.
Chronic cough of consumptive disease, asthma in the
Clinical
aged, short breathing, fever, fullness of chest, perspiration,
Manifestations
night sweat, anorexia.
9 g of small honeyed pills or 1 big honeyed pill, 2 times a
Usage & Dose
day.
Storage Preserve in tightly closed containers.
Specification 9 g per 60 small honeyed pills, 9 g per big honeyed pill.

Figure 2. The typical query result about formula

29
 Chapter 2

TCMID Test Result

Chinese
A wei
Name 
Latin Name Resina Ferulae, 
English
Chinese Asafetida
Name 
Plant Name Ferula sinkiangensi K. M. Shen;Ferula fukanensis K. M. Shen
Properties
Warm,Pungent,Bitter
and Flavors
Meridians Spleen,Stomach,Liver

Toxicity Non toxic

Chinese Asafetida is the resin of Ferula sinkiangensi K. M. Shen or Ferula
fukanensis K. M. Shen (Fam. Umbelliferae). The stem is incised from the
Collection upper part to the lower part at the flowering and early fruiting stages in the
end of spring and early summer. The exuding emulsive resin is collected,
and dried in the shade. 
Irregular lumps or resinous substance. Colour varying in intensity, externally
waxy-yellow to brownish-yellow. Lumps light, texture waxy; fracture
somewhat pored; freshly cut surface pale, gradually darken on storage.
Description
Resinous substance viscid, greyish-white. odour, strongly and lastingly
characteristic and alliaceous; taste, pungent, with the burning sensation on
chewing. 
Function To remove food stagnancy. To disintegrate masses. To kill worms.
To relieve symptoms of stagnation of undigested meat, mass in the
Indication
abdomen due to blood stasis, abdominal pain due to intestinal parasitonis
Therapeutic
Food digestion
Class

Storage  Preserve in well-closed containers, stored in a cool and dry place. 

Usage &
1-1.5 g. 
Dosage 

alpha-pinene, beta-caryophy, lene, beta-eudesmol, beta-myrcene,

Chemical
beta-pinene, camphene, delta(3)-carene, gamma-terpinene, limonene,
Ingredients
myrcene, p-cymene

Figure 3. The typical query result about herb

30
 Chapter 2

TCMID Test Result

Name alpha-pinene, 
CAS RN 2437-95-8.
Molecular Formular C10H16
Molecular Weight 136.2360
A Wei,
Herbs Ai Ye,
Bai Zhi
Structure Download
Proteins
Cytochrome P450 CYP2B1
inhibited/antagonized
Sensory irritation receptor; Increased
Proteins
androstenedione 16alpha-hydroxylase
activated/agonized
activity; Increased aniline hydroxylase activity
Enzymes with
(+)-bornyl diphosphate synthase; 1,8-cineole
compound as
synthase
product
Induction of cytochrome P450s CYP2E1,
Other molecular
CYP2C11, CYP2C6 and CYP6B7; Induction of
events
cytochrome b5 mRNAs

Figure 4. The typical query result about compound

Figure 3 gives the details about the herb, including Chinese Name, Latin Name,

English Name, Plant Source, Properties and Flavors, Meridians, Toxicity, Chemical

Ingredients, Therapeutic Class, Therapeutic Sub-Class, Function, Indications, Clinical

Manifestations, Collection, Processing, Description, Usage and Dosage, Storage and

Literature Reference. The chemical ingredient can link to the corresponding page

containing detail chemical information. They include Chemical Name, CAS No,

molecular weight, molecular formula, downloadable structure file and so on (Figure 4).

31
 Chapter 2

2.4 Data Submission and Updating

This database is to be updated quarterly, information regarding newly searched

TCM formulae, herbs and compounds and additional knowledge them will be added.

To facilitate submission of new information concerning TCM by other users, a data

submission interface is included in the database (Figure 5). Submitted information will

be integrated into the database after validation.

Figure 5. The data submission interface

2.5 Preliminary Analysis of Database

This database currently contains 1343 entries of formulae, 1325 entries of herbs

found from the literature along with 4468 entries of compounds which are extracted

from the herbs. Among these entries of formulae, 1190 entries are relatively complete.

Their herbal components, functions, clinical manifestations are collected into the

32
 Chapter 2

database. However, the others lack one or more detail features, at least the information

are not found by now.

According to their therapeutic effects, the formulae are classified into 17 Classes,

including formulae to release the exterior, purgative formulae, mediation formulae,

formulae for clearing heat, formulae to dispel cold, tonic formulae, astringent formulae,

formulae for tranquilizing the mind, formulae to promote resuscitation, Qi formulae,

blood formulae, formulae for treating wind, formulae for treating dryness, formulae for

dispelling dampness, formulae to expel phlegm, digestive formulae, and formulae for

dispelling parasitic worms. For example, formula Xin Jia Xiang Ru Yin is used to treat

exterior syndromes due to invasion by summer-heat, dampness and cold. It was

classified under ‘formula to release the exterior’ in the database.

Among these entries of herbs, 1090 entries are relatively complete. Their functions,

clinical manifestations are collected into the database. One or more detail features

should be added to the other entries when updating.

According to their therapeutic effects, the herbs are classified into 23 Classes,

including anaesthesia, anthelmintic, antitumor, astriction, bleeding control, blood

activation and stasis removal, food digestion, for calming liver and containing wind,

for dissolving dampness by flavors, for dissolving phlegm, stopping cough, and

soothing breathing, for dispelling wind-dampness, for eliminating toxic materials,

dissolving rottenness and growing new muscles, for promoting diuresis and

penetrating dampness, for relieving exterior syndrome, for tonifying weakness, for

33
 Chapter 2

warming interior, heat clearance, mind opening, purging, regulation of Qi, spirit

calming, toxication reduction, anthelmintic, dampness removal, and itching control,

and vomiting promotion. In addition, each class is divided into several sub-classes.

These classifications will facilitate the researchers in collecting and comparing the

herbs which have the similar effects. For example, Semen Lablab album (Bai Bian Dou)

is used in treatment of weakness of the spleen and stomach with loss of appetite and

loose bowels, excessive leukorrhea, vomiting, diarrhea, distress in the chest and

distension in the abdomen caused by summer-damp. In the database, it belongs to the

Class ‘for tonifying weakness’ and the sub-class ‘for tonifying Qi’.

After further investigation, we found that some herb entries are the same as or very

similar with the other ones. For example, the database contains ‘Ba Yue Zha’ and ‘Yu

Zhi Zi’. These are two different Chinese name of the same herb. A link is provided in

each entry to that of the same herb under different name. Thus, user can get sufficient

information about the herb even if he only knows one name of this herb. Most of the

compounds have the downloadable structure files, which can facilitate the studies of

compound-biomolecule interactions and drug design.

2.6 Conclusion and Further Development

TCMID is developed from available information about formulae, herbs and

ingredients of herbs in the literature and books, which is a result of collective and

persistent effort over the years. It provides a new platform to facilitate mechanistic

34
 Chapter 2

study and clinical evaluation of TCM. In contrast to traditional literature and books,

the database has the following advantages: reduced data redundanc y, data consistency,

easier use of data and less storage for larger information. In addition, new information

can be easily incorporated or the corresponding new databases can be cross-linked to

TCMID to provide more details about TCM.

35
 Chapter 3

Chapter 3: Development of a Computer-aided

Method for TCM Prescription

Formulation

3.1 Introduction

3.1.1 The Principle of TCM Prescription Formulation

A formula of TCM is composed of one or more Chinese medicinal herbs. It is not a

simple arbitrary mixture, but a carefully assembled combination of herbal components

[1,2,16,50]. Each component in a formula has its particular functions. Moreover, they

are mixed together in proper amounts in accordance with specific principles for

combining herbs. When giving a prescription, herbalist doctor must take into account

the characters of each herbal component in the formula, as well as the stage of the

disease and the condition of the patient. According to the different functions of these

ingredients, they are classified to four categories [15], in order of importance, Jun,

Chen, Zuo, Shi.

Jun (King) are one or two herbal ingredients that give the formula its basic

characteristic actions. They may not be the greatest percentage of the formula, but the

most important clinical herbal components. They play the leading role in the treatment

36
 Chapter 3

of the disease. Chen (Minister) are the ingredients that strengthen the Jun’s action

aiming at the principle symptoms. They enhance, augment or broaden the effect of the

Jun. They also play important role in relieving the secondary symptoms. Zuo

(Assistant) are these parts that have different effects in different formula. Firstly, they

can help the Jun and Chen components in their activities and can also address

secondary symptoms, especially the less important symptoms. Secondly, they will

counteract or attenuate any side effects of the first two combinations. Thirdly, they will

modify the energy of the formula closer to neutral, when more potential Jun and Chen

components are used in treatment of serious diseases. Shi (ambassador) are these parts

enable all of the components to be well absorbed, transported and then reach the right

place where they will take effects, such as liver, heart and stomach. They may also be

used as a dispensing agent or give flavor or texture.

The components in a formula are in close cooperation with one another when

exerting their due effects. Theoretically the formulae that have prove n effective are

precisely formulated in accordance with their definite aim in treatment the disease. For

example, Ma Huang Tang (Decoction of Ephedra), which is composed of Herba

Ephedrae (Ma Huang) (9g), Ramulus Cinnamomi (Gui zhi) (6g), Semen Armeniacae

amarum (Ku Xing Ren) (9g), Radix Glycyrrhizae Preparata (Zhi Gan Cao) (6g), has

the main function to promote sweating and release the exterior. In this condition, the

lung is invaded by cold; so it is necessary to use herbs with pungent flavor and warm

property to expel pathogenic factors from the lung. Therefore, Herba Ephedrae is used

as the Jun ingredient to induce perspiration for dispelling pathogenic wind and cold

37
 Chapter 3

from the body and to relieve asthmatic cough by ventilating the lung and smoothing

the flow of vital energy; Gui zhi is the Chen ingredient that aids Herba Ephedrae in

diaphoresis, warms the channels and stimulates menstrual discharge, reinforces yang,

relieves palpitation, and promotes the descending of Qi; Semen Armeniacae amarum

serves as the Zuo ingredient to relieve cough and asthma by ventilating the lung, and to

relax bowels ; and Radix Glycyrrhizae Preparata serves as the Shi ingredient to

harmonize the action of all the other ingredients and direct them to their proper

channels, to reinforce Qi and promote blood circulation. Each formula must contain a

Jun ingredient. However, some formulae may lack one or more of the other categories.

For example, Ren Shen Tang (Decoction of Ginseng) has only one ingredient: Radix

Ginseng (Ren Shen). Its function is to reinforce the vital energy, to remedy collapse

and restore the normal pulse, to benefit the spleen and lung, to promote the production

of body fluid, and to calm the nerves. In some formulae, the Jun or Chen ingredient

may also serve as the Zuo or Shi ingredient. To prescribe an effective formula, the

doctor should have good understanding of the characters of hundreds of herbs.

3.1.2 Modification of TCM Prescription

The composition of a formula is not static. Patients suffering from the same disease

may be given different variations of the same formula. Even the same person, during

different disease stage, is given modified formula. Factors such as the disease

condition, the constitution and age of the patient, the season, and the geographical

environment should be considered when composing a formula [20]. A very famous

38
 Chapter 3

doctor of Qing Dynasty, Xu Lingtai, said in his book [63] that “before trying to use any

old handed-down formula, you must first determine whether all the symptoms and

signs of the patient you are treating are consistent with those described in that

prescription and whether every ingredient is good for the case. If not you must make

the proper modifications or, if it is impossible to do so, choose another one.” From

what he said, we can see that the formula components are sometimes modified. The

Jun ingredient must remain the same in the modified formula, but other ingredients are

added or deleted according to the signs and symptoms present and the stage of the

disease. For example, Formula Ke Xue Fang has the effects of clearing fire and

resolving phlegmand. But if the condition is associated with severe deficiency of Yin

and blood, Radix Ophiopogonis (Mai Men Dong), Radix Glehniae (Sha Shen) and

Radix Trichosanthis (Tian Hua Fen) should be added. If the condition is associated

with severe cough, Radix Asteris (Zi Wan), Flos Farfarae (Kuan Dong Hua) and

Semen Armeniacae (Xing Ren) should be added. If the condition is associated with

Cough with profuse blood, Bai Bu should be added. Another example is Formula Huo

Xiang Zheng Qi San, which is to release the exterior and resolve dampness. If the

condition is associated with Indigestion, Massa Fermentata medicinalis (Shen Qu) and

Semen Raphani (Lai Fu Zi) should be added. While if the condition is no chill, fever or

headache, Radix Angelicae dahuricae (Bai Zhi) and Folium Perillae (Zi Su Ye) should

be deleted. The examples show that it is not always necessary to follow the set formula.

On the contrary, proper variations should be made. Similarly, it is not hard to

understand that their relative amounts are also varied according to the patient’s

39
 Chapter 3

condition. Prescriptions consisting of the same substances in different proportions may

have different effects. Besides, different forms of the same prescription may have

different level of intensities or actions. The major forms of the prescriptions include

decoctions, powders, pills, capsules, and honey boluses. Usually in bolus form, the

action is slow. But when a decoction of the same ingredients is used instead, the action

becomes quicker [20]. Therefore the decoction form is more appropriate than the bolus

in serious or critical conditions.

3.1.3 Previous Studies on Prescription Formulation

In TCM, prescriptions, which are composed of various herbs according to the

principles, are often considered as the major means of treating diseases. Since ancient

times, TCM physicians have summarized their experiences in this field. Many

effective prescriptions have thus been tested and documented. However, people were

disappointed to find that there was no direct and reliable approach to prescribe new

formula toward particular disease. What they can do and are doing is to try to modify

and improve the existed formulae. Moreover, how the formula may be modified is also

depended on their special experiences and the constitution of the patients. Fortunately,

scientists have made progress in developing alternative approach to facilitate the study

of the formulation of new prescription. Employing a fuzzy clustering method, Su

Weiwei et al divided the herbs in one formula into several different parts according to

their characters [39-41]. These characters were quantified to generate the feature

vectors that can be recognized by computer. From the first chapter, we notice that

40
 Chapter 3

different herbs have different characters and functions. That is why different herbs are

used for treatment of different diseases. To understand the mechanism of herbs and use

them properly, it is necessary to explore their characters. Ancient experts have, from

the TCM viewpoint, summarized those basic characters, which include drugs'

Properties and Flavors, Meridians and Toxicity, etc. Based on the theories of Yin and

Yang, Viscera, Channels and Collaterals, treatment principles of traditional Chinese

medicine have been developed and summed up throughout a long history of medical

practice. From the ancient Chinese medicine books, we can also find that the herbs

with similar characters have similar effects, while those with different characters have

quite different effects. In Su’s studies [40,41], they analyze several formulae that have

proven effective, including Wan Shi Niu Huang Qing Xin Wan, Xiao Chai Hu Tang,

Hua Tuo Zai Zhao Wan and Nao De Sheng Pian. The results show that the

classification is in accordance with the traditional explanation of Chinese medicine.

3.2 A New Computer-aided Method for Prescription Formulation

Su’s studies, introduce in the previous chapter, provide us with a method through

which the characters of herbs can be quantified. We have noticed that a TCM

prescription is a selective combination of multiple herbs, which indicates the

probability of building the quantified characters of the prescription as a whole,

according to those of the herbal components. Based on this analysis, we employ

Support Vector Machine (SVM), a machine learning method, for the first time, to

explore the principle of prescription formulation, to evaluate the effectiveness of a

41
 Chapter 3

formula and even generate new formulae.

3.2.1 Support Vector Machine (SVM)

SVM is one kind of learning machine based on statistical learning theory. The

basic idea of applying SVM to pattern classification is as follows: First, the input

vectors are mapped into a feature space, either linearly or non- linearly, which is

relevant with the selection of the kernel function. And then a hyperplane of the vectors

in the feature space is constructed which separates two classes (this can be extended to

multi-class). SVM training always seeks a global optimized solution and avoids

over- fitting, so it has the ability to deal with a large number of features. A complete

description to the theory of SVMs for pattern recognition is in Vapnik’s book [43].

SVMs have been employed in a wide range of real-world problems such as text

categorization [64-66], hand-written digit recognition [67,68], tone recognition [69-71],

image classification and object detection [72-77], sub-storm and flood stage

forecasting [78,79], cancer diagnosis [80-83], glaucoma diagnosis [84], microarray

gene expression data analysis [85-87]. It has been shown that SVM is consistently

superior to other supervised learning methods [88,89].

When used for classification, SVMs separate a given known set of {+1, -1} labeled

training data via a hyperplane that is maximally distant from the positive samples and

negative samples (known as Optimal Separating Hyperplane, OSH), then ‘plot’ the test

data at the high dimensional space, distinguish whether it belongs to positive or

negative according to the OSH. Figure 6 illustrates this process. The solid lines show

42
 Chapter 3

two possible hyperplanes, each of which correctly separates the training data into two

classes. The two dashed lines parallel to the separating hyperplane show the

boundaries in which one can move the separating hyperplane without misclassification.

We call the distance between each parallel dashed lines as margin.

Small Margin Large Margin

Figure 6. Two possible separating hyperplanes

For most of real- world problems, seemingly not to be linearly separable, SVMs can

handle non- linear hypotheses with any Kernel function, which automatically realizes a

nonlinear mapping onto a feature space. The optimal separating hyperplane found by

the SVM in feature space corresponds to a nonlinear decision boundary in the input

space.

3.2.2 Linear Classification

Let the training data of two separable classes with n samples be represented by
G G G G
( x1 , y1 ), ( x 2 , y 2 ), " , ( x n , y n ), i = 1,2," , n, where xi ∈ R N is an N dimensional space,

43
 Chapter 3
G
and yi ∈ {−1,+1} is the class index. Given a weight vector w and bias b (Figure

7), the two classes can be separated by two margins parallel to the hyperplane:

G G
wT • xi + b ≥ 1, yi = +1 (1)

G G
wT • xi + b ≤ −1, yi = −1 (2)

G
where w = ( w1, w2 ,", wn )T is a vector of n ele ments. Inequalities (1), (2) can be

combined into a single inequality:

G G
yi ( wT • xi + b ) ≥ 1, i = 1,2,", n (3)

M
w
Hyperplane

Margin γ K K
w T ⋅ x + b = +1
−b K K
w
K wT ⋅ x + b = 0
K K
wT ⋅ x + b = −1

Figure 7. Definition of Hyperplane and Margin. The circular dots and square dots

represent samples of class -1 and class +1, respectively.

44
 Chapter 3

M H M
H”

H’

Figure 8. Schematic of the available Hyperplanes H, H’, H’’…, about a set of

training data.

OM
OSH

OM

Figure 9. Schematic of unique Optimal Separation Hyperplane of a set of training

data.

45
 Chapter 3

As shown in Figure 8, there exist a number of hyperplanes for each group of

training data. The classification objective of SVM is to determine an optimal weight

G
w0 and an optimal bias b0 such that the selected hyperplane separates the training
G
data with maximum margin. The hyperplane determined by w0 and b0 is called

optimal separation hyperplane (OSH) which is illustrated in Figure 9.

The equation of any given hyperplane can be written as:

G G
wT • xi + b = 0 (4)

and the distance between the two corresponding margins is:

G G G G
G xT • w xT • w
γ ( w, b) = Gmin G − { max
G G (5)
{ x | y = +1} || w || x | y = −1} || w ||

The OSH can be found by maximizing the above distance or by minimizing the norm
G
of w under inequality constraints (3), and

G 2
γ max = γ ( w0 , b0 ) = G (6)
w0

The saddle point of the following Lagrangean gives solutions to the above

optimization problem:

G GT G n GT G
L( w, b,α ) = 12 w • w − ∑ α i[ yi ( w • xi + b) − 1] (7)
i =1

where α i ≥ 0 are Lagrange multipliers. The solution of this optimization

G
Quadratic Programming (QP) problem requires that the gradient of L( w, b, α ) with

G
respect to w and b vanishes, which gives the following conditions by the calculation

46
 Chapter 3

∂L ∂L
of G = 0 and = 0:
∂w wG =wG 0 ∂b b =b 0

G n
G
w0 = ∑ α i yi xi (8)
i =1

∑α y i i =0 (9)
i =1

By substituting Eq.(8) and (9) into Eq.(7), the QP problem becomes

maximization of the following expression:

n n n
G G
L(α ) = ∑ α i − 12 ∑∑α iα j yi y j ( xiT • x j ) (10)
i =1 i =1 j =1

n
under the constrains ∑α y i i = 0 and α i ≥ 0, i = 1,2,", n.
i =1

The points situated at the two optimal margins have non-zero coefficients α i

among the solutions to Eq.(10), and are called Support Vectors (SV). The bias b0 can

be calculated as follows:

G G G G
b0 = − 12 ( G min w0T • xi + Gmax wT0 • xi ) (11)
{ xi | y i = +1} { x i | yi = −1}

After determination of support vectors and bias, the decision function that

separates the two classes can be written as:

n
GT G GT G
f ( x ) = sign[ ∑α i yi xi • x + b0 ] = sign[∑ α i yi xi • x + b0 ] (12)
i =1 SV

3.2.3 Nonlinear Classification

As real- world problems are usually nonlinear, the following approach has been

47
 Chapter 3

G
introduced into SVM to deal with these problems. The original training data x in the

input space X is projected into a high-dimensional feature space F via a Mercer kernel

operator K [67], and the OSH is constructed in this feature space. Such a procedure is

illustrated in Figure 10.

In mathematical terms, the set of classifiers is transformed into the form:

G G
f ( x ) = sign[ ∑α y K ( x , x ) + b ]
i∈{ SV }
i i i 0 (13)

where K is a symmetric positive definite function, which satisfies Mercer’s conditions,

∞
G G G G
K ( x, y) = ∑α m φ ( x ) T • φ ( y ), αm ≥ 0
m =1

G G G G G G G G
∫∫ K ( x , y ) g ( x ) g ( y ) dxdy > 0, ∫g ( x )dx < ∞
2
(14)

The Kernel represents a legitimate inner product in input space:

G G G G
K ( x , y ) = φ ( x )T • φ ( y ) (15)

In the F-space, the dual Lagrangian, given in Eq.(10), is

n n n
G G n
L(α ) = ∑α
i =1
i − 1
2 ∑∑α α
i =1 j =1
i j yi y j K ( xi , x j ) − λ ∑α i yi
i =1
(16)

n
subject to ∑α y i i = 0 and α i ≥ 0, i = 1,2,", n.
i =1

and the decision function is:

G G
f ( x ) = sign[ ∑α y K ( x , x ) + b ]
i∈{ SV }
i i i 0 (17)

where

48
 Chapter 3

G G G G
b0 = − 12 { Gmin (
{ x i | yi = +1}
∑α
j ∈{ SV }
j y j K ( xi , x j )) + Gmax (
{ xi | y i = −1}
∑α
j ∈{ SV }
j y j K ( xi , x j ))} (18)

X F φ(x)
X X
φ(o) φ(x)
X O
φ φ(x)
O
O φ(o) φ(x)
O
O φ(o)
X X φ(o) φ(x)
φ(o)

Figure 10. Illustration of basic principle of support vector machines: Step 1:

project the training data nonlinearly into a higher-dimensional feature space via φ .

Step 2: construct a hyperplane to separate positive and negative datasets with

maximum margin there.

A number of candidate kernel functions have been used in SVM, including

G G2
G G G G G G x− y
Polynomial K ( x , y ) = (1 + x • y ) d , Gaussian RBF K ( x , y ) = exp( − ) ,
2σ 2
G G
G G x− y
Exponential RBF K ( x , y ) = exp( − ) , and their comb inations by summing
2σ 2

kernels or tensor products of kernels [90]. In this work, Gaussian RBF is chosen as the

kernel function.

49
 Chapter 3

3.3 Dataset Preparation

In this study, 644 TCM formulae (from TCMID database, Chapter 2) are used to

test the feasibility of this method. Based on the number of herbs in a formula, these

formulae are classified to 7 groups, from Group 1 to Group 7, in which the number of

herbs is 4, 5, 6, 7, 8, 9 and 10, respectively. These formulae are used as positive

samples. 1000 negative samples, which contain the same number of herbs with the

positive samples in the same group, are randomly generated. These negative samples

are supposed to be therapeutically noneffective because they are only randomly

compiled complexes of multiple herbs. The total numbers of samples in each group are

given in Table 3. Both positive and negative formulae in each group are further

randomly divided into two parts with even number of entries, respectively (If the

number of the positive formulae is an odd number (2n+1), they are divided into two

parts with n and n+1 formulae, respectively.). One part is used as the training set,

while the other as the testing set. Thus, from Group 1 to Group 7, the training sets

contain 42, 41, 46, 42, 50, 47 and 55 positive samples, respectively and 500 negative

samples which are randomly selected. The testing sets contain 42, 40, 46, 42, 50, 46

and 55 positive samples, respectively and 500 negative samples.

3.4 Feature Vectors

Construction of feature vectors is one of the key steps in successful SVM

classification. Because a formula is composed of several different herbs, at first we

50
 Chapter 3

must construct the vectors of each herb, and then, construct the feature vectors of the

formula according to some rules that can compile the vectors of all herbs. In this study,

the feature vectors are simply the concatenation of the vectors of all herbs. I will

explain it in the following paragraphs.

Table 3. Number of Positive Formulae and Negative Formulae in Each Group

Formula Number of Number of Positive Number of Negative

Group Herbs in a Formulae Formulae

Formula

Training Testing Training Testing

Set Set Set Set

Group 1 4 42 42 500 500

Group 2 5 41 40 500 500

Group 3 6 46 46 500 500

Group 4 7 42 42 500 500

Group 5 8 50 50 500 500

Group 6 9 47 46 500 500

Group 7 10 55 55 500 500

51
 Chapter 3

Table 4. Principle for constructing the feature vector

Elements Characters Value

1 Cold 1 (Cold), 1.2 (Extremely cold), or 0.8 (Slightly cold) *

2 Hot 1 (Hot) or 1.2 (Extremely hot) *
3 Warm 1 (Warm), 1.2 (Extremely warm) or 0.8 (Slightly warm)
*
4 Cool 1 (Cool) or 0.8 (Slightly cool) *
5 Pungent 1 (Pungent) or 0.8 (Slightly Pungent)
6 Sweet 1 (Sweet) or 0.8 (Slightly Sweet) **

7 Sour 1 (Sour) or 0.8 (Slightly Sour) ***

8 Bitter 1 (Bitter) or 0.8 (Slightly Bitter)
9 Salty 1 (Salty) or 0.8 (Slightly Salty)
10 Toxic 1 (Toxic), 1.2 (Extremely Toxic), or 0.8 (Slightly Toxic)
11 Lung 1 (if has)

12 Bladder 1 (if has)

13 Spleen 1 (if has)
14 Large Intestine 1 (if has)
15 Stomach 1 (if has)
16 Small Intestine 1 (if has)

17 Liver 1 (if has)

18 Pericardium 1 (if has)
19 Heart 1 (if has)
20 Kidney 1 (if has)
21 Gallbladder 1 (if has)
22 San Jiao 1 (if has)

* If the herb has mild property, the value of each element should be added 0.25.

** If the herb has tasteless flavor, the value of this element should be added 0.5

*** If the herb has astringent flavor, the value of this element should be added 0.5.

52
 Chapter 3

Since the characters of a herb are essential to its effects, the feature vectors should

be constructed according to these characters. The characters are quantified and then the

values of them are used to compose the feature vector. The vector contains 22 elements,

which are the quantified basic characters of herbal properties, flavors, toxicity property

and meridians that the herb can enter. Table 4. gives probable value for each element of

the feature vector and also the reason for the choice of the value.

As we have known from previous chapters, each herb has its own properties and

flavors. There are four typical types of properties, that is, cold, hot, warm or cool.

These four characters are used as four elements of feature vector. If the herb has one of

these properties, a value ‘1’ is assigned to the element. Otherwise, a value ‘0’ is

assigned to this element. The choice of value ‘1’ is to simplify the construction of the

vector.

There is also another property: mild. If the herb has this property, the value of the

elements, which represent Cold, Hot, Warm and Cool, add a value 0.25. It is because

this property does not belong to any of these four typical ones. The value 0.25 is

chosen mainly according to our experiences. So are the other values, which are chosen

to describe the elements in the following paragraphs. The discussion of the choice of

the value is given in Chapter 3.6. Some herbs are extremely cold and some extremely

hot, the value ‘1.2’ is assigned to the elements, which represent cold and hot,

respectively. To those herbs with minor cold, minor warm or minor cool, a value ‘0.8’

is assigned to the corresponding element.

53
 Chapter 3

Table 5. Example: Feature Vector of Herba Ephedrae (Ma Huang)

Elements Characters Value

1 Cold 0
2 Hot 0
3 Warm 1
4 Cool 0
5 Pungent 1

6 Sweet 0
7 Sour 0
8 Bitter 0.8
9 Salty 0
10 Toxic 0

11 Lung 1
12 Bladder 1
13 Spleen 0
14 Large Intestine 0
15 Stomach 0

16 Small Intestine 0
17 Liver 0
18 Cardiovascular 0
19 Heart 0
20 Kidney 0
21 Gallbladder 0
22 San Jiao 0

54
 Chapter 3

There are five major types of herbal flavors, including pungent, sweet, sour, bitter

and salty. These characters are also used as elements of the feature vector. If the herb

has one of these flavors, a value ‘1’ is assigned to this element. If the flavor is minor

pungent, minor sweet, minor sour, minor bitter or minor salty, a value ‘0.8’ is assigned

to the corresponding element. There are also another two different flavors: tasteless

and astringent. For tasteless, a value ‘0.5’ is added to the element ‘sweet’ and for

astringent, to ‘sour ’. According to Viscera, Channels and Collaterals theory, the

symptoms can reflect the organs that are not in good conditions. So ancient Chinese

herbalist doctors speculated that a given herb may selectively act upon a particular part

of the body and this part depends on the corresponding symptoms that the herb can

relieve. So the vector should also reflect the properties of meridians that herbs affect.

There are a total of 12 meridians, including Lung, Bladder, Spleen, Large Intestine,

Stomach, Small Intestine, Liver, Pericardium, Heart, Kidney, Gallbladder and San Jiao.

The quantified characters of these meridians are also used as elements of the feature

vector. If one herb can enter a meridian, then a value ‘1’ is assigned to the element that

represents for the meridian. ‘0’ is assigned to the other elements, which are not related

with this herb. Table 5. gives the feature vector of Herba Ephedrae (Ma Huang), a

useful herb which has warm properties, pungent and slightly bitter flavors, and can

enter Lung and Bladder meridians. The related information can be found in TCMID

database. So does the information of the other herbs, on which the feature vectors are

built.

After building the feature vector of each herb in a formula, we can construct that of

55
 Chapter 3

the formula. In this study, I connect the elements of all herbs’ vector together to

generate a bigger vector. For example, the formula, which contains 4 herbs, has the

feature vector with 88 elements, while the vector of a formula (contains 5 herbs) has

110 elements. Because of the difference of feature vector, we separated the formulae

into 7 groups and tested them individually (Table 3).

3.5 Accuracy Measure

Accuracy of the results from discriminative methods is commonly measured by the

quantity of True Positives (TP), True Negatives (TN), False Positives (FP), and False

Negatives (FN) [91,92]. In addition to these quantities, standard sensitivity, specificity

and overall accuracy (Q) performance measures defined by

Sensitivity=TP/(TP+FN) (19)

Specificity=TN/(TN+FP) (20)

TP + TN
Q= (21)
TP + TN + FP + FN

are also useful in assessing the prediction accuracy [93]. To estimate performance, we used

three objective indices: accuracy, specificity and sensitivity. The sensitivity as defined

above is the probability of correctly predicting a positive example. The specificity as

defined above is the probability of correctly predicting a negative example. All these

quantities are used in the evaluation of SVM classification of formulae in this work.

56
 Chapter 3

3.6 Results and Discussion

The SVM program can automatically rearrange the training and testing sets.

During the training process, the program can calculate the minimum number of the

positive samples or negative samples in the training set, which is necessary to reach an

optimal training result. If the number of the positive samples or negative samples in

the training set is higher than the minimum, the rest samples are automatically

transferred to the testing set. On the contrary, if the number is lower than the minimum,

some samples in the testing set are taken into the training set and then re-start the

training process until an optimal training result is achieved. Tables 6 to12 give the

training and testing sets of positive formulae of each groups after the optimal training

results are achieved and Table 13 summarizes the number of positive and negative

samples in the training set and testing set of each group after the training process. The

results show that the number of the positive samples in the training sets of these groups

increase (see Table 3). This is perhaps because the numbers of positive samples in the

original training set is not enough to cover sufficiently diverse range of positive

samples. So a number of positive samples in initial testing set are put into the training

set. On the other hand, except Group 1, in the other six groups, the number of the

negative samples in the training set decrease from 500 to a certain number between

210 and 290. The decrease of the number of negative samples in the training set

indicates that they are enough to achieve the optimal result.

57
 Chapter 3

Table 6. List of Positive Formulae in the Training and Testing Set of Group 1

List of positive formulae (contain 4 herbs)

Training Set

Bai Tou Weng Tang Huang Teng Nan Chang Sheng Ma Huang Qi Tang
Bao Chi San Tang Shi Zao Tang
Bei Xie Fen Qing Yin Huo Xue Tong Mai Pian Si Jun Zi Tang
Bing Lang Chen Qi Tang Ji Jiao Li Huang Wan Si Miao Yong An Tang
Bing Peng San Jia Wei Dang Gui Bu Xue Si Ni San
Cai Shi Tui Re Tang Tang Si Sheng Wan
Cang Er Zi San Jian Shen Fang Si Wei Huang Qi Xuan Fu
Da Cheng Qi Tang Jiu Fen San Pian
Da Jian Zhong Tang Ling Gui Zhu Gan Tang Si Wei Tu Mu Xiang San
Da Xian Xiong Wan Lou Xing Chen Qi Tang Si Wu Tang
Dao Chi San Ma Huang Tang Tao Hua San
Ding Xiang Shi Di Tang Ma Huang Xing Ren Yi Yi Tong Xie Yao Fang
Ding Zhi Wan Gan Cao Tang Wei Jing Tang
Fu Zi Xie Xin Tang Ma Xin Shi Wei Tang Wu Long Jian Ji
Gan Cao Gan Jiang Ling Ma Xing Gan Shi Tang Wu Zhu Yu Tang
Zhu Tang Meng Shi Gun Tan Wan Xi Jiao Di Haung Tang
Gan Mao Tui Re Chong Ji Mu Fang Ji Tang Xiang Su San
Gan Mao Tui Re Ke Li Qian Yin Tang Jiang Xiao Er Qi Ying Wan
Gui Zhi Gan Cao Long Gu Qian Zheng San Xiao Zhi Shuan
Mu Li Tan Qing Bing San Xie Bai San
Gun Tan Wan Qiong Yu Gao Xin Shen Yin
He Zi Pi San Re Yan Ning Ke Li Xuan Bai Cheng Qi Tang
Hong Yao Tie Gao San Ao Tang Yang Xue Run Chang Jian
Huai Hua San San Liang Ban Yao Jiu Yi Nian Jin
Huang Lian Jiang Tang San San Ye Tang Yu Quan Wan
Huang Lian Jie Du Tang Shen Fu Tang Zhi Ke Pi Pa Tang Jiang
Huang Qin Tang Sheng Ma Ge Gen Tang

Testing Set

Bai Hu Tang Ge Gen Huang Qin Huang Li Zhong Wan

Er Chen Wan Lian Tang Ling Gui Zhu Gan Tang
Fu Ling Gui Zhi Bai Zhu Ge Gen Qin Lian Pian Ma Xing Shi Gan Tang
Gan Cao Tang Ge Gen Qin Lian Wei Wan Ge Gen Huang Lian Tang

58
 Chapter 3

Table 7. List of Positive Formulae in the Training and Testing Set of Group 2

List of positive formulae (contain 5 herbs)

Training Set

An Dan Tang Jian Gu Tang Tai Yi Gao

Bai Dai Wan Jian Xin Fu Mai Lin Tao Ren Cheng Qi Tang
Bai Hu Jia Gui Zhi Tang Jing Zhi Guan Xin Pian Tian Jin Gan Mao Pian
Ban Xia Hou Po Tang Ju Yuan Jian Tian Jin Zhi Ke Tang Jiang
Bao Long Wan Kui Yang Pian Tong Mai Yang Xin Wan
Bi Xie Fen Qing Yin Ling Bao Hu Xin Dan Wen Pi Tang
Bu Shen Qiang Shen Pian Ma Xin Long Xie Xiong Wu Ling San
Da Huang Mu Dan Tang Tang Wu Pi San
Dan Qi Tang Nao De Sheng Pian Wu Pi Yin
Dang Gui Zhi Tong Tang Niao Shi 1 Hao Wu Shi Tang
Fu Kang Ning Pian Ning Sou Lu Xie Huang San
Fu Tu Dan Qing Hao Bie Jia Tang Xin Jia Xiang Ru Yin
Fu Zheng Gu Ben Fang Qing Hou Yan He Ji Yi Huang San
Fu Zi Da Huang Tang Qing Nao Jiang Ya Tang Yi Huang Tang
Fu Zi Li Zhong Wan Qing Wei San Yin Pu Xiao Du Yin
Gua Lou Qu Mai Wan Qing Xuan Wan Yu Nu Jian
Guan Xin Su He Wan Qing Yi 1 Hao Yue Bi Tang
Gui Zhi Fu Ling Wan Qu Tao Tang Yue Ju Wan
Gui Zhi Fu Zi Tang Run Chang Tang Zeng Ye Cheng Qi Tang
Gui Zhi Tang Sang Xin Shi Gao Tang Zhen Jiang Gao Yao
Hou Po Sheng Jiang Ban Shen Xue Pian Zhen Wu Tang
Xia Gan Cao Ren Shen Sheng Bai Chong Ji Zhi Ke Pi Pa Chong Ji
Tang Shi Qu Gan Ce Da Zao Zhu Ling Tang
Huan Xin Dan Tang Zhu Sha An Shen Wan
Huang Lian E Jiao Tang Si Fo He Ji Huo Xue Chu Xuan Tang
Huang Long Tang Suan Zao Ren Tang

Testing Set

Tao He Cheng Qi Tang Gui Zhi Jia Gui Tang Huang Qi Gui Zhi Wu Wu
Bai Hu Jia Ren Shen Tang Ling Gan Wu Wei Jiang Tang
Bao Yuan Tang Xin Tang Kang Bai Huo He Ji
Fu Fang Da Qing Ye He Ji

59
 Chapter 3

Table 8. List of Positive Formulae in the Training and Testing Set of Group 3

List of positive formulae (contain 6 herbs)

Training Set

An Shen Ding Zhi Wan Jie Gu Gao Sheng Hua Tang

Ban Xia Lu Jie Re He Ji Sheng Yi Xuan Hua Tang
Bei Mu Gua Lou San Jin Ji Chong Ji Si Jun Zi Wan
Bing Lang Si Xiao Wan Jin Pu Tang Si Shen Wan
Bu Fei E Jiao Tang Jin Suo Gu Jing Wan Suo Pi Yin
Chan Su Ding Jin Yin Tang Tao Hong Si Wu Tang
Chuan Xiong Zhi Tong Tang Ku Shen Tong Lin Tang Tou Nong San
Da Bu Yin Wan Lai Zhi San Tuo Hua Jian
De Sheng Wan Lian Mei An Hui Tang Wan Shi Niu Huang Qing Xin
Er Cha Er Huang San Liang Di Tang Wan
Er Chen Tang Ling Yang Gan Mao Pian Wu Mei Chen Qi Tang
Er Zhu Ling Pi Tang Liu Shen Wan Wu Tou Gui Zhi Tang
Fei Lian Fen Qing Tang Liu Wei Di Huang Wan Wu Wei Xiao Du Yin
Fu Fang Gan Yan Chong Ji Ma Pu San Xi Huang Wan
Gou Teng Yin Ma Ren Run Chang Wan Xiao Er Xiao Shi Pian
Gu Jing Wan Mai Men Dong Tang Xiao Huo Luo Dan
Guan Xin 2 Hao Yao Nao Ling Su Pian Xiao Jian Zhong Tang
Gui Fu Yi Lin Tang Ping Wei San Xiao Tong Tang
Hua Ban Tang Qi Shu Hua Fen Tang Xu Shi Tong Jing Fang
Hua Shui Zhong Zi Tang Qin Lian Pian Xuan Jiang Tang
Huo Xue Zhi Tong San Qing Gong Tang Yi Guan Jian
Ji Chuan Jian Qing Ying Jie Yu Tang Yi Mu Cao Gao
Jia Liu Wan Ru Kuai Xiao Pian Yi Qi Ju Xian Tang
Jia Wei Jiao Hong Yin Run Chang Wan Yi Yi Zhu Ye San
Jia Wei Si Ni San San Cai Feng Sui Dan Yin Chai Chong Ji
Jiang Can Er Huang San Shan Zha Hua Zhi Wan Yin Qiao Tang
Zuo Gui Yin Shen Zhu San Yue Bi Jia Shu Tang

Testing set

Ai Hu Cheng Qi Tang Fu Ling Pi Tang Jia Wei Fu Xie Ning

Bai Mai An Shen Yin Gui Fu Li Zhong Wan Jia Wei Wu Yao Tang
Fang Ji Huang Qi Tang Huo Luo Dan Ju Pi Zhu Ru Tang
Feng Re He Ji Dang Gui Shao Yao San Li Huang Tang

60
 Chapter 3

Table 9. List of Positive Formulae in the Training and Testing Set of Group 4

List of positive formulae (contain 7 herbs)

Training Set

Ba Zhen Yi Mu Wan Jia Jian Dang Gui Si Ni Shan Zha Xiao Mi Tang
Ban Long Wan Tang Shang Dong E Jiao Gao
Ban Xia Xie Xin Tang Jia Wei Gui Zhi Fu Ling She Xiang Zhen Tong Gao
Bao He Wan Wan She Yang Tang
Bu Fei Shen Ji Tang Jia Wei Zhi Zhuo Gu Ben Shen Jin Dan Jiao Nang
Bu Yang Huan Wu Tang Tang Sheng Ge Er Chong Tang
Cai Hu Qing Gan Yin Jiang Ya Ping Pian Shun Jing Tang
Chai Hu Shu Gan San Jie Biao Tui Re Fang Tian Tai Wu Yao San
Chong Cao Bu Tian Jing Jin Lian Mao Gong Yin Tiao Gan Tang
Da Qing Long Tang Ku Shen Tang Wan Ying Gao
Da Yuan Yin Lan Di Tang Wei Kang Ling Jiao Nang
Dang Gui Liu Huang Tang Li Pi Su Fei Tang Xiao Chai Hu Tang
Dang Gui Si Ni Tang Lian Po Yin Xiao Er Yi Niao Fang
Di Yu Wan Liu Yi Bing Zhu San Ma Xiao Yao Wan
Fei Er Wan You Hu Xing Jun San
Fu Fang Chan Yi Yin Luo Han Guo Zhi Ke Chong Xuan Fu Dai Zhe Tang
Fu Fang Lei Gong Teng Ji Yang He Tang
Jian Ji Ma Zi Ren Wan Yang Wei Li Qi Tang
Gan Cao Xie Xin Tang Mi Zhen Tang Yi Qi Huo Xue Pian
Gan Ji San Mu Li Ze Xie San Yu Zhen San
Ge Gen Tang Qi Bao Mei Ran Ke Li Yun Qi San
Gong Wai Yun Fang Qi Wei Du Qi Wan Zhen Xin An Shen Tang
Hong Ling San Qi Wei Pu Tao San Zhi Sou San
Hua Gan Jian Qi Wei Zhi Shen Tang Zhi Zhi Tang
Huang Lian Tang Qian Bai Bi Yan Pian Zhu Ye Mai Dong Shi Gao
Huang Qin Hua Shi Tang Qiang Huo Sheng Shi Tang Tang
Huang Tu Tang Qin Lian Si Wu Tang Zhu Ye Shi Gao Tang
Ji Ming San Qing Li Tang Sha Shen Mai Dong Tang
Sang Xing Tang Qu Feng Zhi Tong Pian

Testing Set

Dang Gui Si Ni Tang Dang Gui Jian Zhong Tang Huang Qi Jian Zhong Tang
Huo Bu San Ren Tang Ji Xue Teng Gao

61
 Chapter 3

Table 10. List of Positive Formulae in the Training and Testing Set of Group 5

List of positive formulae (contain 8 herbs)

Training Set

Ba Wei Qing Xin Chen Xiang Kuo Qing Yin Si Chong Huo Xue Tang
San Li Gan Long Pian Snag Piao Xiao San
Ba Wei Tan Xiang San Li Ling Tang Su Wei Tang
Ba Zhen Wan Ling Yi Bai Jiang Tang Su Xiao Niu Huang Wan
Ban Xia Bai Zhu Tian Ma Mi Niao Xi Jie Shi Fang Tan Du Jian
Tang Nu Bao Tong Jing Zhi Ke Tang
Bei Ling San Qi Gong Tang Tong You Tang
Bi Xue Shen Shi Tang Qi Guan Yan Wai Tie Fang Tuo Li Ding Tong Tang
Bu Dai Wan Qi Ju Di Huang Wan Wen Dan Tang
Bu Zhong Yi Qi Wan Qi Li San Wu Cao Tang
Chan Su Wan Qi Shao Ji Cao Tang Wu Jia Qiang Xin Tang
Chuan Bei Qing Fei Tang Qiang Xin Ying Xian Qi Yin
Jiang Qing Gan He Jie Tang Xiang Sha Liu Jun Zi Wan
Chuan Xiong Cha Tiao Wan Qing Gu San Xiao Ban Tang
Er Long Zuo Ci Wan Qing Guo Wan Xiao Qing Long Tang
Fu Yuan Huo Xue Tang Qing Lin He Ji Xiao Yao San
He Che Da Zao Wan Quan Long Si Zi Tang Xie Qing Wan
Hua Yu Qing San Tang Ren Shen Ding Chuan Tang Xie Shi Tang
Hui Tian Zai Zao Wan Rou Gan Xi Feng Tong Luo Yang Yin Li Yan Tang
Huo Xue Run Zao Sheng Jin Tang Yang Yin Qing Fei Tang
San Ru Xiang Ding Tong Wan Yi Chan He Ji
Jia Jian Tuo Hua Jian San Bai Tang You Gui Yin
Jia Jian Wei Rui Tang San Qi Shang Yao Pian Zao Du Fu Jiang Tang
Jia Wei Ban Xia Hou Po Tang San Ren Tang Zhen Xin Di Tan Tang
Jia Wei Fu Zi Li Zhong Tang San Shen Tang Zhi Shi Dao Zhi Wan
Jia Wei Xiao Yao San Sang Ju Yin Zhi Shu He Wei Tang
Jian Wei Cha Shen Kang Ning Pian Zhi Xue Tang
Jiao Ai Tang Shen Qi Wan Zuo Gui Wan
Jie Nue Qi Bao Yin Shu Gan Chong Ji Kang Bao Kou Fu Ye
Ku Shen Zhi Yang Tang Kai Xiong Shun Qi Wan
Testing Set

Da Bu Yuan Jian Gou Zai Hua He Ji Juan Bi Tang

Da Chai Hu Tang Hou Po Wen Zhong Tang Mai Wei Di Huang Wan
E Jiao Yi Shou Jing Huang Qi Gui Di Tang Niu Huang Jie Du Pian
Er Dong Tang Jia Wei Wu Pi Yin Niu Huang Jie Du Wan
Er Yin Jian Jin Gui Shen Qi Wan Nuan Gan Jian
Fu Fang Yi Bai He Ji

62
 Chapter 3

Table 11. List of Positive Formulae in the Training and Testing Set of Group 6

List of positive formulae (contain 9 herbs)

Training Set

Ai Fu Nuan Gong Wan Ju Fang Zhi Bao San Ti Qi Yi Zhong Tang

An Kun Zan Yu Wan Jun Li Tang Tiao Wei He Ji
Ba Zhen San Kai Wei Qi Cao Ting Tao Chen Qi Tang
Chuan Xiong Cha Tiao San Kang Gu Zeng Sheng Wan Tong Qiao Huo Xue Tang
Da Zao Wan Kang Yuan Tang Wei Hu Tang
Ding Tong Tang Li Shi Hua Yu Tang Wei Qi Shuang Jie Tang
Ding Xuan Tang Liang Ge San Wei Yan Yan Fang
Fu Fang She Dan Chuan Ling Yang Qing Fei San Wen Yang Tong Mai Tang
Bei San Lu Jin Yin Wu Shao She Yin
Geng Nian An Mu Dan Pi Tang Xi Ling Jie Du Pian
Gu Ci Wan Niu Bang Jie Ji Tang Xiao Er Jin Dan Pian
Gu Yin Jian Pai Shi Chong Ji Xiao Tan Hua Jie Tang
Gui Ling Gan Lu San Qi Long Tang Xiao Ying Tang
Gui Ling Ji Qi Zhen Wan Xie Fei Tiao Zhong Fang
Gui Zhi Shao Yao Zhi Mu Qing Li An Shen Tang Xuan Bi Tang
Tang Qing Qi Hua Tan Wan Xuan Fei Zhi Ke Tang
Hu Qian Wan Qing Xin Lian Zi Yin Xue Xie Gua Lou Tang
Huang Qi Yi Shen Tang Qing Yan Wan Yang Xue Dang Gui Jing
Huang Qi Zhu Fu Tang Qing Ying Tang Yi Jia Jian Zheng Qi San
Jia Jian Yi Qi Gu Chong Qing Zao Jiu Fei Tang Yi Qi Yang Yin Tang
Tang Ren Shen Bai Du San Yin Qiao Jie Du Wan
Jia Kang Ping Wan Ren Shen Zi Bu Gao Yu Yin Zhi Beng Tang
Jia Wei Xiao Yao Wan Ru Mi Niao Tang Zhi Gan Cao Tang
Jian Pi Ding Xian Tang Sha Mai Yu Tian Tang Zhi Shi Xiao Pi Wan
Jiao Mei Bing Guan Tang Shao Fu Zhu Yu Tang Zhi Zhuo Gu Ben Wan
Jin Fei Cao San Shao Yao Tang Zhu Shi Tang
Jing Zhui Fang She Gan Ma Huang Tang Zhu Ye He Ji
Jiu Wei Qiang Huo Keli Shen Jiang Xie Xin Tang Zi Shen Rong Jing Wan
Jiu Wei Qiang Huo Tang Sheng Li Pian Zi Yin Xuan Fei Tang
Jiu Wei Qiang Huo Wan Si Ling He Qin Shao Tang Jiu Xian San
Jiu Wei Zi Shen Wan

Testing set

Jia Jian Bu Zhong Yi Qi Jiu Wei Xiao Ding Yin Ma Xin Qin Ting Tang
Tang Ding Chuan Tang Man Gan Jiu Wei Tang

63
 Chapter 3

Table 12. List of Positive Formulae in the Training and Testing Set of Group 7

List of positive formulae (contain 10 herbs)

Training Set
An Nao Niu Huang Pian Jia Wei Xiang Su San Shu Jin Huo Xue Pian
An Shen Bu Xin Wan Jian Nao An Shen Tang Shu Zao Yin Zi
Bai He Gu Jin Wan Jie Du Huo Xue Tang Su Zi Jiang Qi Wan
Bao Gu Jiu Jie Xiao Tang Tai Shan Pan Shi San
Bu Gan Yi Qi Tang Kui Yang He Ji Tang Shen Yi Tang
Bu Shen Yi Nao Tang Lan Wei Xiao Yan Pian Tian Ma Jiao Nang
Can Shi Tang Li Dan Pai Shi Pian Tian Ma Wan
Da Huang Qing Wei Wan Lian Kui Yu Yang Tang Wan Dai Tang
Da Huo Luo Wan Ling Jiao Gou Teng Tang Wei Chang Fu Yuan Tang
Dan Dao Pai Shi Er Hao Lu Shang Yu Zhen Wei Tong San
Fang Lu Tai Gao Wen Fei Hua Yin Tang
Dan Zhi Xiao Yan San Mi Yuan Jian Wen Jing Huo Xue Tang
Fang Nian Lian Tang Mie Di Ku Shen Tang Wen Yang Yi Qi Tong Mai
Fu Fang Zang Lian Wan Mie Di Tang Tang
Fu Gan Tang Mu Xiang Shun Qi Wan Wu Mei Tang
Gan Lu Yin Niu Huang Bao Long Wan Wu Mei Wan
Gan Wei Bai He Tang Qing Fei Yi Huo Wan Xi Xian Gou Ji Xian Lin Pi
Gong Ti Wan Qing Fei Yin Zi Tang
Gu Chong Tang Qing Gan An Tai Yin Xiao Ji Yin Zi
Han Lian Cao Jian Qing Shi Yi Qi Tang Xiao Jin Dan
Hou Zheng Wan Qu Feng Zheng Rong Tang Xiao Yan Shen Ji He Ji
Hu Gu Yao Jiu Re Du Qing Xin Jia Huang Long Tang
Hu Po Bao Long Wan San Huang Shi Gao Tang Xin Shuai He Ji
Hua Yu Zhi Beng Tang Shang Gan Tang Yi Qi Chu Tan Fang
Hui Yan Zhu Yu Tang Shao Fu Zhu Yu Wan Yin Jun Hu She Tang
Huo Xiang Zheng Qi Kou Fu Shen Lu Bu Gao Yin Qiao San
Ye Shen Rong Jiu Zhen Ren Yang Zang Tang
Huo Xiang Zheng Qi Shui Shi Hu Si Jin Tang Zhi Chuang Xun Xi Fang
Huo Xue Tong Luo Tang Shi Quan Da Bu Gao Zhi Dai Fang
Ji Sheng Shen Qi Wan Shi Quan Da Bu Wan Zhou Che Wan
Ji Xue Teng Tang Shi Wei San Zhu Huang Bei Mu Tang
Jia Jian Xiao Cai Hu Tang Shu Gan Huo Xue Tang Zhuang Yao Jian Shen Wan
Jia Wei Bai He Di Huang Tang Jia Wei Jiao Ai Si Wu Tang
Testing Set
Ba Zhen Tang Dao Chi Wan Jian Nao Bu Shen Wan
Bu Zhong Yi Qi Tang Fu Ke Shi Wei Pian Long Dan Xie Gan Tang
Cai Hu Yu Jin Pai Shi Tang Hua Shuan Tong Mai Tang Long Dan Xie Gan Wan
Chen Xiang Hua Qi Wan Jia Jian Zhu Qi Bu Tai Tang Ren Shen Gui Pi Wan
Dang Gui Yang Xue Wan Jia Wei Yin Qiao San Bai He Gu Jin Tang
Jiu Qi Nian Tong Wan

64
 Chapter 3

Table 13. Number of Samples in the Training and Testing sets after Calculation Using
SVM

Formula Training Set Testing Set

Group

Positive Negative
Positive Negative
TP FN TN FP

1 74 501 10 0 498 1

2 73 251 5 3 749 0

3 80 211 10 2 786 3

4 79 267 1 4 733 0

5 84 231 10 6 768 1

6 88 290 3 2 710 0

7 94 210 14 2 790 0

Description of TP, FN, TN and FP in Chapter 3.5

The prediction accuracy of SVM in classifying TCM formulae is given in Table 14.

The ove rall accuracy of SVM for classifying Group 1 was 99.8%; the sensitivity, 100%;

the specificity, 99.8%. The overall accuracy of SVM for classifying Group 2 was

99.6%; the sensitivity, 62.5%; the specificity, 100%. The overall accuracy of SVM for

classifying Group 3 was 99.4%; the sensitivity, 83.3%; the specificity, 99.6%. The

65
 Chapter 3

overall accuracy of SVM for classifying Group 4 was 99.5%; the sensitivity, 20%; the

specificity, 100%. The overall accuracy of SVM for classifying Group 5 was 99.1%;

the sensitivity, 62.5%; the specificity, 99.9%. The overall accuracy of SVM for

classifying Group 6 was 99.7%; the sensitivity, 60%; the specificity, 100%. The overall

accuracy of SVM for classifying Group 7 was 99.7%; the sensitivity, 87.5%; the

specificity, 100%.

Table 14. Sensitivity, Specificity and Overall Accuracy of the Results

Formula Herbs Sensitivity Specificity Q

Group Number
TP/(TP+FN) TN/(TN+FP) (TP+TN)/(TP+TN+FN+FP)

Group 1 4 100% 99.8% 99.8%

Group 2 5 62.5% 100% 99.6%

Group 3 6 83.3% 99.6% 99.4%

Group 4 7 20% 100% 99.5

Group 5 8 62.5% 99.9% 99.1%

Group 6 9 60% 100% 99.7%

Group 7 10 87.5% 100% 99.7%

Description of Sensitivity, Specificity, Overall Accuracy, TP, FN, TN and FP in Chapter 3.5

66
 Chapter 3

The high overall accuracies (>99%) may reflect the reliability of this method. At

least, it has strong ability to recognize negative samples. The specificities of all 7

groups are bigger than 99.5%. For Group 2, 4, 6 and 7, all of the negative samples are

recognized. Thus this SVM system appears to be capable of minimizing the probability

of incorrect selection of non-effective formulae. However, the high overall accuracies

may not truly reflect its ability to pick up “true positive ”, that is, the known effective

formulae. In this study, the values of overall accuracies largely depend on the number

of “true negative ” samples, which is much higher than “true positive” samples. That is

the reason why, in some groups, Sensitivity is much lower than Specificity and Q. In

other words, we can say that recognition ability of “true positive ” is different in

different groups, or, more seriously, the evaluation of a known effective formula may

be incorrect. Table 14 shows satisfactory results for Group 1, Group 3 and Group 7

(sensitivity > 83%). These groups have a common character, that is, in the testing set,

the positive samples (TP and FN) are relatively more than, at least not less than those

samples in the other groups. In Group 1, Group 3 and Group 7, the positive sample is

10, 12 and 16, respectively. While in Group 4, the recognition ability is particularly low,

with only 20% of the formulae correctly classified. The number of the positive testing

samples in this group is 5. Similarly, in Group 6, the number of the positive testing

formulae is also 5. For this group, the sensitivity is 60%. Considering the small number

of these positive testing samples, the results in these groups may be not reliable.

Several factors that limit this method are given in the next paragraphs.

67
 Chapter 3

Several factors may affect the prediction accur acy. Firstly, the number of positive

samples is critical to the predicted results. As shown in Table 14, the lower values of

sensitivity than those of specificities may be the consequence of lack of positive data.

Secondly, the number of negative samples also needs careful consideration. This

number should be higher than, but not much higher than the minimum. In this study,

the number of the negative testing samples is too high, while that of the positive testing

samples is too low. Further studies should be carried out to reduce the number of the

former and to find more positive samples for testing. Moreover, SVM prediction can

be further improved by choosing better SVM optimization procedure and feature

vector. The selection of feature vector may be more important. This includes the

selection of the value of each element and the construction rule of formulae’s feature

vector from herb’ feature vector. Changing to a proper value of an element may

generate reasonable results. For example, when a herb is extremely cold, the element

that represents “cold” is assigned a value 1.2 (Table 4). This value can be changed to

1.1, 1.2 or whatever, and then a new feature vector is generated. Based on a set of these

new feature vectors, an optimal training result may be achieved. This will be carried

out in the future study. There may be a number of possible methods to construct the

feature vector of the formulae from the herb’s feature vector. In this study, I select a

simple one (see Chapter 3.4). This method has an obvious disadvantage, that is, it can

only train the formulae that contain the same number of herbs. Hence, I have to

separate the formulae into several groups and then train and test them respectively. The

other methods for constructing the feature vector sho uld be employed in the future

68
 Chapter 3

study.

Table 15. False Predicted Negative Formulae (or Potential Formulae)

Herb Formula Name Herbal Components Description

Number

4 Negative Formula-4 443 Mu Xiang, Chuan Xin Lian FP

Ye, Zhi Ma Qian Zi, Shuan
Hua
6 Negative Formula-6 18 Man Jin Zi, Ji Nei Jin, Deng FP
Xin Cao, Duan Long Gu,
Man Jin Ye, Xi He Liu
6 Negative Formula-6 414 Fa Ban Xia, Cang Zhu, Guo FP
Lu Huang, Zhi Cao, Meng
Shi, Zi Bei Chi
6 Negative Formula-6 481 Chui Pen Cao, Ye Ming FP
Sha,Ya Tuo Cao, Jiu Da
Huang, Ting Li Zi, Shan
Zha

8 Negative Formula-8 533 Zhi Jun Tan, Xiao Gan FP

Cao,Gan Song, Su Geng, Su
He Xiang You, Rou Cong
Rong, Tu Ku Shen, A Wei

Table 15 gives the ineffective formulae that are incorrectly predicted as effective

formulae. One possible reason for the incorrect prediction is the inherent deficiency of

this method as described above. But we cannot rule out the possibility that these

randomly generated formulae do have effective in treatment of some kinds of diseases.

In conclusion, our study suggests the capability of SVM in recognizing

non-effective formulae and it may provide some helpful hints for herbalist doctors to

determine the effectiveness of a TCM formula. In addition, the computation provides

69
 Chapter 3

several potentially effective formulae from the hundreds of randomly mixed formulae.

These formulae never appear in previous literature and their effects have never been

studied before. It is unclear whether these formulae have the therapeutic value. The

method is expected to facilitate the prescription of new and novel TCM formulae as

well as the validation of existing TCM formulae while more and more formulae are

under scientific studies.

70
 Chapter 4

Chapter 4: Exploration of the molecular mechanism

of a Medicinal Herb Serenoa repens by

INVDOCK

4.1 Introduction

The pharmacological use of phytotherapeutic agents in the treatment of benign

prostatic hyperplasia (BPH) and associated lower urinary tract symptoms (LUTS) has

been growing steadily [45-49,94-96]. The most widely used phytotherapeutic agent is

the extract of Serenoa repens, one kind of dwarf palm plants [45-49,97-104]. Many

clinical trials have been carried out to evaluate the effect of the extract of this herb

[98,102,103,104,105]. A recent meta-analysis of published clinical trial data of

Permixon, one commercial product of the extract of Serenoa repens, (11 randomized

clinical trials and 2 open label trials) [100], involving 2859 patients, reveals that in

men with symptomatic BPH, the extract yields a significant improvement in peak flow

rate and reduction in nocturia compared with placebo. Studies in hormone-treated

castrated rats also show inhibition effects on the increase of prostate wet weight by the

extract of Serenoa repens [107]. When compared with commonly used drugs, an

alpha-blocker (Tamsulosin) [98] or 5 alpha-reductase inhibitor (Finasteride) [108], the

extract shows nearly the same effects, causing parallel and statistically significant

71
 Chapter 4

decreases in symptom scores and increases in maximal flow rates. These results

indicate the effects of the extracts of Serenoa repens on men with symptomatic BPH.

Its main mechanism of action is still unclear; however, possible activities have

been demonstrated by several in vivo and in vitro studies, including antiandrogenic,

antiproliferative and anti- inflammatory activities [96]. 5alpha-reductase catalyzes the

reduction of testosterone into dihydrotestosterone (DHT), which is critical in

regulating prostate cell growth. Studies demonstrate that the lipido-sterol extract of

Serenoa repens (LSESr) markedly inhibit 5alpha-reductase [109-111]. In addition, it

also exerts an anti-androgenic effect through inhibition of the receptor binding of

androgens [111].

In one study, LSESr significantly inhibits proliferation and induces cell death in

both epithelium and stroma of BPH issue [112]. It also appears to affect the

b-FGF-induced proliferate of prostate cells [113] and decreases the amount of EGF in

BPH [105]. Another study demonstrates that Serenoa repens may block PRL-induced

prostate growth by inhibiting several steps of PRL receptor signal transductions [104].

It exerts an anti- inflammatory effect through the inhibition of the enzyme responsible

for prostaglandin and leukotriene synthesis, such as cyclooxygenase and lopoxygenase

[105].

The mechanism of this herb is unknown. It is also unclear which of the compounds

in the extract of this herb account for its therapeutic effects. Many experiments assume

that the lipid and sterol in this herb account for the therapeutic effects

72
 Chapter 4

[101,103,104,107,109,112-117]. Shimada et al separates two compounds from the

extract: 1-monolaurin and 1-monomyristin, and finds that they show moderate

biological activities in the brine shrimp lethality test and against renal (A-498) and

pancreatic (PACA-2) human tumor cells [118]. A number of compounds have been

extracted from this herb. To explore the mechanisms and pharmacology of this herb,

these ingredients need to be examined first. And it is more important to explore how

they can collectively interact with various proteins and other molecules in the human

body. But such an investigation is hindered by the high cost and time spent on

synthesis and bioassay of natural products. Therefore alternative methods are needed

to understand and analyze molecular mecha nism of active ingredients at a higher speed

and lower cost.

As one of the strategies for mechanistic study of natural products, the software

INVDOCK is developed to calculate the binding between the compounds and the

protein targets [119]. To evaluate the capability of INVDOCK in identification of

potential therapeutic and toxicity protein targets of compounds, Chen et al has tested

on a more diverse set of compounds [42,120]. The predicted potential therapeutic and

toxicity protein targets for a number of clinical drugs were found to be in fair

agreement with available experimental data. Knowledge of these potential targets

combined with that of proteomics and pharmacokinetic profile can facilitate the

assessment of therapeutic and toxic effects as well as the molecular mechanism of

predicted effects.

When this work was done, I did not have the 3D structure of 5 alpha-reductase and

73
 Chapter 4

androgen receptor, so I did not evaluate the effects of the compounds on these two

targets. In this work, I focus on the mecha nism of this herb involved in other processes

to provide the experimentalists the most possible targets, which, I believe, will

facilitate them in the construction of experiments. 23 active compounds from this herb

are selected to search for their potential targets through INVDOCK. Because these

compounds are the main ingredients of Serenoa repens, the collective effects of them

may account for parts of the actions of this herb in the treatment of BPH.

4.2 INVDOCK Method

4.2.1 Protein Cavity Database

According to Kuntz’ s work [121], a cavity is modeled by a group of overlapping

spheres that fill up that cavity. Each cavity entry is derived from the corresponding

PDB entry by the following procedure: Ligands and water in a PDB protein structure

are first removed. The surface of this protein, as defined by Richards [122], is then

generated. The van der Waals surface of a solvent-accessible atom is generated with

the respective parameter from the AMBER force field [123]. The inward- facing

surface covering the interface of van der Waals surfaces is computed with a probe

sphere 1.4 Å in radius. The whole protein surface is then coated and filled with a

cluster of spheres by a method similar to that of Kuntz et al. Each sphere is checked for

the extent of its surrounding space covered by protein atoms. The surrounding space is

defined as a region within 15 Å of the center of the sphere. A sphere is considered

covered if more than 50% of the direction around the sphere is covered by protein

74
 Chapter 4

atoms. The covered spheres are then divided into separate cavity groups on the basis of

the spatial separation of nearest neighboring spheres.

In addition, non-covered spheres close to a covered sphere in a group are included

in that group. This is to ensure that a cavity model sufficiently covers the cavity

surface region where a tail of a ligand might be located. In some ligand-protein PDB

structures, such as HIV-1 protease complex with its inhibitor, one or both ends of the

ligand are found to be located in such a region. The computer-generated model

includes both the cavity and the groove to which it is connected. Because a groove

may also be a binding site, no attempt has been made to remove grooves from a cavity

model. A visual inspection of these and a large number of othe r proteins showed that

all the cavities and their adjacent grooves were reasonably represented.

3D structures of proteins from different species have been deposited in the PDB.

Because of species-related variation in protein sequences, ligand binding is often

specific to a protein of a particular species, or a species class, and those of close

phylogenetic distance. In general, the main interest of target identification is to

understand the molecular mechanism of the therapeutic and physiological effect of a

ligand with respect to one species or a few particular species. Therefore, cavity entries

can be divided into classes defined by disease-related species, sequence similarity, and

phylogenetic distance to a particular species or species class such as human and

mammal. Additional information such as the known ligand-binding region and the

75
 Chapter 4

extent of each sphere being buried inside a cavity is also included in a cavity entry to

facilitate the prioritized docking strategy described next.

4.2.2 Inverse-docking Procedure

The inverse-docking procedure INVDOCK is designed for the automated search of

every entry of the protein cavity database for potential protein targets of a small

molecule. Because of the large number of cavity entries, across-the-board specification

of possible binding sites within each cavity is difficult. Hence, all parts of each cavity

are subject to docking. Docking to sites inside each cavity starts from known

ligand-binding sites, followed by more interior sections and then the remaining part.

To save central processing unit (CPU) time, the program proceeds to the next protein

entry when the first successful dock is obtained without an exhaustive search for the

optimum binding mode within each cavity. Although the optimum binding mode is no t

specifically sorted, the prioritized search strategy seems to dock a ligand to a site

reasonably close to the experimentally observed positions in most of the ligand-protein

structures studied. All cavity entries are subject to searching unless the related protein

has been identified as a potential target.

Flexible ligand docking is similar to the multistep strategy approach proposed by

Wang et al [124] Ligand conformation is sampled at a resolution similar to that of

Wang et al. The docking of a particular conformer to a cavity is as follows: First, the

ligand is aligned within the selected site by the position of each ligand atom being

matched with the center of the spheres. Because of the relatively low-resolution nature

76
 Chapter 4

of ligand conformation sampling, a certain degree of structural clash is allowed at this

stage. A molecular- mechanics conformation optimization is then conducted by a

limited torsion space sampling of rotatable bonds in the ligand and those in the

side-chain of the receptor amino acid residues at the binding site. Each rotatable bond

is sampled at ±15°. This is followed by 50 iterations of Cartesian coordinate energy

minimization on all ligand and protein atoms at the binding site to further optimize the

ligand-protein complex. Energy minimization is by a steepest decent method.

In both torsion optimization and energy minimization, AMBER force fields [123]

are used for covalent-bond, bond-angle, torsion, and nonbonded van der Waals and

electrostatic interactions. A large number of crystal structures in PDB contain only

non-hydrogen atoms. To save computing time and avoid the difficulty in modeling

hydrogens, the Morse potential [125], which is a function of the donor-acceptor

distance, is used to represent hydrogen-bond terms. This potential has been shown to

give a reasonable description of hydrogen-bond energy and dynamics in biomolecules

[126,127]. The published Morse potential parameters [126,127] are used in this work.

Protein flexibility is known to influence ligand binding, and thus methods such as

side-chain conformational sampling and protein ensemble generation have been

explored in docking studies [128,129]. Limited side-chain conformation sampling is

considered in INVDOCK along with additional energy minimization for all atoms at

the binding site. However, no attempt is made to explicitly sample the ensemble of

protein conformations. A substantial number of proteins in PDB have multiple entries

77
 Chapter 4

submitted by different groups; some of them are associated with different binding

ligands or different mutants. Moreover, there are multiple conformations in most NMR

structures. To a certain extent, INVDOCK searching of these multiple forms of

structures/conformations may serve as a partial sampling of the conformation for some

proteins.

4.2.3 Scoring

The scoring of docked structures is based on a ligand-protein interaction energy

function, ?ELP , composed of the same hydrogen-bond and nonbonded terms as those

used for structure optimization. An analysis of a large number of PDB ligand-protein

complexes shows that the computed ? ELP is generally less than ? ET hreshold = -αN

kcal/mol, where N is the number of ligand atoms and α is a constant (∼1.0). The exact

value of α can be determined by the fitting of the linear equation ?EThreshold = -αN to

the computed ?ELP for a large set of PDB structures. This statistically derived energy

value can be used empirically as a threshold for screening likely binders. A polynomial

form of ? EThreshold involving more parameters may also be introduced to derive an

energy threshold. ?ELP can be required to be less than ?EThreshold when successfully

docked structures are selected.

Ligand binding is competitive in nature. A drug is less likely to be effective if it

binds to its receptor noncompetitively against natural ligands and, to some extent,

other drugs that bind to the same receptor site. This binding competitiveness may be

partially taken into consideration for cavities known to be ligand-bound in at least one

78
 Chapter 4

PDB entry. Ligands in PDB structures are known binders. Therefore, PDB ligands

bound to the same receptor site as that of a docked molecule may thus be considered

competitors of that molecule. In INVDOCK selection of a potential protein target, the

computed ? ELP is not only evaluated against ? EThreshold but also compared to the

ligand-protein interaction energy of the corresponding PDB ligands that bind in the

same cavity in this or other relevant PDB entries. The ligand-protein interaction energy

for the relevant PDB structures is computed by the same energy functions as that for

the docked molecule. In addition to the condition that it be lower than ?Ethreshold, ?ELP

of a docked molecule is required to be lower than a competitor energy threshold

? ECompetitor when a potential target is selected. ?ECompetitor can be taken as the highest

ligand-protein interaction energy of the corresponding PDB ligands multiplied by a

factor β. For finding weak binders as well as strong binders, a factor β≤1 is introduced

to scale the ligand-protein interaction energy of PDB ligands. This is because a weak

binder may have a slightly higher interaction energy than that of a PDB binder. We

have not found experimental data to determine the value of β. Here, β is tentatively

determined by an analysis of the computed energy for a number of compounds. Our

study suggests that a value of 0.8 for β leads to reasonable results statistically.

4.2.4 Selection of Compounds and Therapeutic and Toxicity Proteins

23 ingredients of Serenoa repens (Table 16) are collected according to published

literature and books [96,130-132]. The 3D structures of these ingredients are generated

through the following steps. First, the 2D structures are got from the Combined

79
 Chapter 4

Compound Database (CCD) (http://www.chemnetbase.com). Second, these structures

are converted to 3D structures through the software Weblab. The structures are given

in Figures 11 to 14.

Therapeutically targeted proteins are well documented in pharmacology literature

and in a number of publications. A database of therapeutic targets, TTD

(http://xin.cz3.nus.edu.sg/group/TTD), has been developed [133]. TTD currently

contains 433 protein entries that have been reported to be the target of either clinical

drugs or investigating drugs.

Table 16. Herbal ingredients of Serenoa repens

Compound Class Compound Name

Fatty acid Capric acid, Caproic acid

(8) Caprylic acid, Lauric acid
Myristic acid, Oleic acid
Palmitic acid, Stearic acid
Sterol Beta-sitosterol, Cycloartenol

(5) Lupenone, Lupeol, Stigmasterol

Ethyl ether of fatty Ethylcappate, Ethylcaproate, Ethylcaprylate,
acid Ethyllaurate, Ethyloleate, Ethylmyristate,
(8) Ethylpalmitate, Ethylstearate,

Monoacylglycerides Monolaurin, Monomyristin

(2)

80
 Chapter 4

Figure 11. 3D Structure of Phytosterols of Serenoa repens

A: Beta-sitosterol, B: Cycloartenol C: Lupenone, D: Lupeol, E: Stigmasterol

Figure 12. 3D Structure of Monoacylglycerides of Serenoa repens

A: Monolaurin, B:Monomyristin

81
 Chapter 4

Figure 13. 3D Structure of Fatty acids of Serenoa repens

A: Caproic acid, B: Caprylic acid, C: Capric acid, D: Lauric acid, E: Myristic acid, F:
Oleic acid, G: Palmitic acid, H: Stearic acid

82
 Chapter 4

Figure 14. 3D Structure of Ethyl Esters of Fatty acids of Serenoa repens

A: Ethyl caproate, B: Ethyl caprylate, C: Ethyl cappate, D: Ethyl laurate, E: Ethyl
myristate, F: Ethyl oleate, G: Ethyl palmitate, H: Ethyl stearate

83
 Chapter 4

4.3 Results

All the 23 ingredients, including eight fatty acids, eight ethyl esters of these fatty

acids, five sterols and two monoacylglycerides, are used to scan the protein library to

identify their potential targets. Based on the information, their contribution to the

therapeutic and toxic effects of Serenoa repens is evaluated (Table 16). These

compounds constitute more than 85% of all the ingredients in the extract of Serenoa

repens, which is used for treatment of BPH and is believed to contain the

pharmacologically active components. Although their mechanism of action is still

unclear, theoretically, the therapeutic effects of this herb may be due to some of these

compounds.

From the computational results, each compound can bind to several protein targets,

ranging from one to seven. Taking into account of the overlap of the same targets by

different compounds, finally, 25 protein targets are obtained (Table 17 and Table 18),

each of which can be bound by seve ral different compounds. To a given protein target,

the binding energy of different compounds varies significantly. For example, the

results show that 20 compounds can bind to prostaglandin H2 synthase (the data is not

included in the thesis), but their binding energy is very different. The highest one is

that of stigmasterol and the lowest one is that of capric acid. The energy is 47.6

kcal/mol (Table 17) and 28.4 kcal/mol (this data does not include in the thesis),

respectively. Compared with the former, the latter is a very weak binder and may have

little contribution to the interaction with the target protein. Hence, in this study, I only

consider the interactions with sufficiently strong binding energy. When the binding

84
 Chapter 4

energy of a compound to a given protein is 3 kcal/mol less than the maximum, the

interaction between the compound and the protein is neglected. The compounds that

can bind to the targets are given in Table 17 and Table 18.

The action of the herbal extract has been suggested to be through the following

mechanism [45,49,100,134]: anti- inflammatory, direct inhibition of prostate growth

and inhibition of growth factor- induced growth, anti-androgenic and anti-estrogenic

effects. A number of proteins involved in these processes are in the list of our predicted

targets.

4.3.1 Anti-inflammatory Effects

The anti- inflammatory effects are mainly through interference with prostaglandin

synthesis [45,100,103]. One critical enzyme involved in this process, Prostaglandin H2

synthase, is found highly exp ressed in prostate cancer and BPH issue, suggesting an

important role in the pathogenesis of BPH. Studies find its expression can be decreased

by the extract of Serenoa repens [135]. One study suggests phytosterols to be the most

importance of inhibition of this enzyme [104]. From our study data, two phytosterols

(stigmasterol and cycloartenol) can bind to this enzyme. On the other hand,

beta-sitosterol, believed to be the most important ingredient, is found not to bind to this

enzyme. Further study need to be carried out to compare their effects.

85
 Chapter 4

Table 17. Predicted Proteins related with BPH. [energy >= (Max-3) kcal/mol]

PDB Protein Name Pathway Ref. Ligand and Energy

ID

1ebv Prostaglandin H2 Prostaglandin [135] Stigmasterol -47.6

1diy Synthase synthesis Cycloartenol -47.2
(Cyclooxygenase)

1fgk FGF Receptor 1 FGF-induced cell [136,138] Stigmasterol -44.8

proliferate
1e0o Fibroblast Growth FGF-induced cell [137] Lauric acid -36.0
Factor 1 proliferate Monomyristin -34.2
Capric acid -33.7
1exz Stem Cell Factor GF-induced cell [138] Stearic acid -43.8
proliferate
7odc Ornithine Polyamine [141,147] Stearic acid -33.7
Decarboxylase synthesis and cell Oleic acid -33.1
growth Myristic acid -31.3

1jen S-Adenosylmethioni Polyamine [142] Palmitic acid -38.3

ne Decarboxyla synthesis and cell Monolaurin -37.7
growth Caprylic acid -37.5

Protein Kinase C Proliferation of [143] Betasitosterol -46.2

stromal cell Monolaurin -46.2
Oleic acid -44.5
Stearic acid -43.7
2hdh L-3-Hydroxyacyl Generation of [146] Oleic acid -37.4
COA Dehydrogenas DHT Lauric acid -36.8

1err Estrogen Receptor Estrogenic [96,106] Stigmasterol -53.5

pathway betasitosterol -46.6
2nll Retinoic Acid Proliferation by [145] Stigmasterol -49.7
Receptor retinal

1ed4 Nitric Oxide NO pathway [144] Monolaurin -47.9

Synthase Lupeol -45.8
Stigmasterol -45.7

86
 Chapter 4

4.3.2 Anti-proliferate Effects

In BPH, bFGF concentration is 2 to 3- fold higher than in normal prostate.

Moreover, over-expression of FGFR1 is observed in BPH [136-138]. Studies also

demonstrate that the extract of this herb can alter bFGF growth factors- induced growth

and proliferation [113]. In their study, it is mentioned that lupenone and the

unsaponified fractio n of the LSESr markedly inhibit the b-FGF-induced cell

proliferation. The inhibition effect may be through reducing the available growth factor

receptor and growth factor. Our data show that one sterol (stigmasterol) can bind to

fibroblast growth factor receptor. Moreover, fibroblast growth factor (FGF) can be

bound by two fatty acid (lauric acid and capric acid) and monomyristin.

Another growth factor, stem cell factor, which is supposed to exert a broad range of

biological activities during organogenesis and normal cell development [139], is

expressed in Prostate carcinoma cell lines: DU-145 and PC-3 lines [140]. Out data

show it is bound by stearic acid.

Besides the inhibition of the growth factor-induced growth of prostate cell,

inhibition of other proteins that are critical to the growth of cells may account for the

effect of this herb. Ornithine decarboxylase (ODC) is essential for polyamine synthesis

and cell growth. Polyamine is believed to participate in cellular proliferation and

differentiation. It has been found that not only the expression level of ODC mRNA,

but also the activity of this enzyme is higher in BPH tissue than that of normal prostate

tissues [141]. Our data show that three fatty acids, stearic acid, oleic acid and myristic

87
 Chapter 4

acid can bind to this enzyme. Another rate- limiting enzyme of the polyamine pathway,

S-adenosylmethionine decarboxylase, is also in the list of predicted targets, binding by

palmitic acid, caprylic acid and monolaurin. Both of them are found to be useful

targets in the treatment of prostate cancer [142]. Protein kinase C (PKC) is involved in

the proliferative response of human cultured prostatic stromal cells [143]. Our data

show that betasitosterol, monolaurin, oleic acid and stearic acid can bind to this

enzyme.

One predicted protein target is nitric oxide synthase. This enzyme catalyzes the

production of NO, which is involved in several important pathways. One study finds

that inducible nitric oxide synthase is expressed in the prostate of all benign prostatic

hyperplasia (BPH) patients, suggesting the role of NO in the pathogenesis of BPH

[144]. Monolaurin and two sterols (lupeol and stigmasterol) can bind to this enzyme.

Vitamin A (retinol) and its derivatives, the retinoids, have been implicated as

chemopreventive and differentiating agents in a variety of cancers, including that of

the prostate. Very little is known about the physiological role of retinoids in the

prostate. Another study finds that the nuclear retinoic acid receptors are expressed in

BPH samples [145]. Inhibition of this receptor may neutralize the proliferate effect

caused by retinal. Two fatty acids (oleic acid and Caprylic acid) and one sterol

(stigmasterol) can bind to this receptor.

4.3.3 Anti-androgenic and Anti-estrogenic Effects

Many studies suggest the anti-androgenic and anti-estrogenic effects of Serenoa

88
 Chapter 4

repens in the treatment of BPH. L-3-hydroxyacyl-CoA dehydrogenase (SCHAD) has

been demonstrated to convert 5alpha-androstane-3alpha, 17beta-diol (3alpha-adiol) to

dihydrotestosterone (DHT) [146]. Dihydrotestosterone (DHT) is the major androgen

implicated in the pathogenesis of BPH. A number of studies have shown that the

Serenoa repens extract takes effect through reducing the generation of DHT [109-111].

It is believed that part of the anti-androgenic effects of this herb is due to the inhibition

of SCHAD. Our data show that two fatty acids (oleic acid and lauric acid) can bind to

this enzyme.

In regard to the anti-estrogenic effects of Serenoa repens, it has been found that it

can competitively block the translocation of cytosolic estrogen receptors to the nucleus.

[96,106]. Two sterols (stigmasterol and betasitosterol) can bind to this receptor. One in

vivo study demonstrates that beta-sitosterol is capable of producing a weak estrogenic

effect only at the lowest dose [148].

Our study suggests that effects of this herb on other targets may be involved in the

treatment of this disease. In the list of our predicted proteins, besides the targets that

have been mentioned previously, there are still some other targets. So, the next task of

our study is to analyze these proteins to make clear the relationship of these proteins

with BPH. Among these proteins, many proteins are found to be involved in cell

proliferating, cell cycle, tumor me tastasis and other process (Table 18). These proteins

had higher expression in BPH or prostate cancer tissue. Targeting toward them can at

least inhibit the conversion from BPH to advanced prostate diseases.

89
 Chapter 4

Table 18. Other predicted important proteins. [energy >= (Max-3) kcal/mol]

PDB Protein Name Pathway or Ref. Ligand and Energy

ID Function
1hcl Cyclin- Dependent Cell cycle [149] Betasitosterol –48.1
Kinase 2 Lupenone -47.1
Lupeol -47.1
Cycloartenol -46.3
1cws Cdc25 B-Type Tyrosine Cell cycle [150,151] Caproic acid -21.2
Phosphatase

1doa GTP-Binding Protein Cell cycle [152,153] Monolaurin -45.2

(CDC42)

1dow Alpha-Catenin Cell cycle [154,155] Capric acid -37.0

Caprylic acid -33.7
1a7c Plasminogen Activator Inhibition of [162] Cycloartenol -49.2
Inhibitor apoptosis Stigmasterol -47.0

1the Cathepsin B Invasion and [156] Stigmasterol -45.9

metastasis Monomyristin –44.7
1e5t Prolyl Endopeptidase High activity in [157] Ethyloleate -33.2
tumor of prostate
1b3d Stromelysin-1 Metastasis [158] Stigmasterol -49.5
(MMP 3) Cycloarteno l -46.5

Neutrophil Collagenase Metastasis [158] Caprylic acid -38.8

(MMP 8)

Fibroblast Collagenase Metastasis [158] Betasitosterol -48.7

(MMP1)

MMP-13 Metastasis [158] Stigmasterol -43.7

Adp-Ribosylation Metastasis [159] Caprylic acid -24.2

Factor 6 Caproic acid -22.2

Chymase Metastasis [160] Monolaurin -43.4

1ext Tumor Necrosis Factor Increase in PC [161] Ethyllaurate -25.1

Receptor Caproic acid -22.8
Ethylcaprylate -22.8

90
 Chapter 4

4.3.4 Arrest of Cell Cycle

Abnormalities of cell division occur in the carcinogenesis, due to overexpression of

positive regulators of cyclin-dependent kinase (CDK) function. Many drugs have

developed toward CDK to induce growth arrest of the cells at G1 phase [149]. Studies

find that the extract of this herb can induce apoptosis and inhibit cell proliferation of

BPH tissues from patients [112]. In our data, four phytosterols (Beta-sitosterol,

lupenone, lupeol and cycloartenol) can bind to this enzyme.

The other two important molecules involved in the cell cycle, Cdc25 B-Type

Tyrosine Phosphatase [150,151] and CDC42 [152,153] may be the potential targets.

From our data, they are targets of caproic acid and monolaurin, respectively.

Alpha-catenin is important in cell adhesion in the process of proliferate. It has been

shown that its production is elevated in prostatic carcinoma in comparison with benign

prostatic hyperplasia and normal prostate [155]. Our data show two fatty acids (capric

acid and caprylic acid) can bind to it. Plasminogen activator inhibitor 1 (PAI-1) is

suggested by one study to be involved in the tumor growth through the anti-apoptotic

effect in the benign human breast epithelial cell line MCF-10A [162]. Our data show

that two phytosterols (cycloartenol and stigmasterol) can bind to this protein.

Awad AB et al [162] finds that phytosterol, especially beta-Sitosterol, inhibit cell

growth and induce cell cycle arrest using MDA-MB-231 human breast cancer cells to

evaluate the effects of phytosterols. Connolly JM’s study shows inhibition effects of

fatty acids on DU145 human prostate cancer cell growth in athymic nude mice [164].

91
 Chapter 4

4.3.5 Anti-metastasis

Matrix metalloproteinases (MMPs) are a family of zinc endopeptidases involved in

tissue remodeling and thus in various disease processes including tumor invasion and

joint destruction. Correspondingly, these enzymes represent attractive targets for

inhibitor design and drug development. In our predicted list, four types of MMPs are

found: MMP1 (Fibroblast Collagenase), MMP3, MMP8 (Neutrophil collagenase) and

MMP13. Their concentrations are significantly higher in PCa patients with metastases

[158]. MMP1 can be bound by beta-sitosterol, MMP3 by stigmasterol and cyc loartenol,

MMP8 by caprylic acid and MMP13 by stigmasterol. ADP-ribosylation factor 6 plays

significant roles in coordinating the transition of epithelial cells from a stationary to a

motile state when metastasis [159]. Caprylic acid and Caproic acid can bind to this

protein. Chymase is first found in mast cells and seems to be involved in the

degradation of the extracelluar matrix (directly and/or by activation of matrix

metalloproteases), stimulation of submucosal gland cell secretion, and

complement- mediated inflammation. Given the variety of pathophysiological

processes in which chymase is involved, inhibitors of this enzyme may be potential

drugs in cardiovascular and inflammatory diseases [160]. Monolaurin can bind to this

protein. Cathepsin B is among the candidate proteinases believed to participate in

invasion and metastasis. Cathepsin B often presents with higher amounts in malignant

tumors than in normal tissues or benign tumors. Numerous clinical studies have

confirmed a correlation between cathepsin B expression, disease progression, and

92
 Chapter 4

clinical outcome for patients with diverse types of tumors [156]. Stigmasterol and

monomyristin can bind to this protein. Prolyl oligopeptidase activity is significant

higher in tumours of prostate, lung and sigmoid, than in the healthy tissues. It supports

the possible involvement of prolyl oligopeptidase in the renin-angiotensin system and

in the pathogenesis of hypertension [157]. Out data show that ethyloleate bind to it.

Some studies have demonstrated the inhibitory effects of fatty acids [165,166] and

phytosterols [163,167,168] on metastasis of cancer cells. The effects of other

compounds, such as ethyloleate, should be investigated further. Although few

experiments have been carried out to prove the relations hip between these proteins and

the pathogenesis of BPH, our study seems to indicate that they play an important role

in this process.

4.4 Discussion

The results show that most of these chemical ingredients of Serenoa repens can

bind to a number of targets, which has been proven by experiments. Moreover, our

computational results also provide additional potential protein targets of these

compounds. The results show in Table 19 that a compound can only bind to one or a

few protein targets, which indicates the role of this herb in the treatment of BPH is due

to a collective effect of the compounds.

The results indicate that not all compounds are helpful to the overall action of this

herb. 5 of the 23 compounds, including ethylcappate, ethylcaproate, ethylmyristate,

93
 Chapter 4

ethylpalmitate and ethylstearate, may have little contribution to the expected actions of

this herb. These five compounds have their own target proteins. But compared with the

other compounds, they are relatively weak binders to these targets. For example, both

ethylstearate and stigmasterol are predicted to interact with Prostaglandin H2 Synthase,

but the latter strongly docked into the protein. Hence the effects of ethylstearate can be

neglected.

Table 19. Summary of Compounds and their predicted targets

Compound Predicted Targets

Capric acid Fibroblast Growth Factor 1
Alpha-Catenin
Caproic acid Tumor Necrosis Factor Receptor
Adp-Ribosylation Factor 6
Cdc25 B-Type Tyrosine Phosphatase
Caprylic acid S-Adenosylmethionine Decarboxyla
Alpha-Catenin
Neutrophil Collagenase (Mmp 8)
Adp-Ribosylation Factor 6
Lauric acid Fibroblast Growth Factor 1
L-3-Hydroxyacyl Coa Dehydrogenas
Myristic acid Ornithine Decarboxylase
Oleic acid L-3-Hydroxyacyl Coa Dehydrogenas
Protein Kinase C
Ornithine Decarboxylase
Palmitic acid S-Adenosylmethionine Decarboxyla
Stearic acid Ornithine Decarboxylase
Protein Kinase C
Stem Cell Factor
Betasitosterol Fibroblast Collagenase (Mmp1)
Estrogen Receptor
Protein Kinase C
Cyclin- Dependent Kinase 2
Next page

94
 Chapter 4

Continued
Compound Predicted Targets
Cycloartenol Prostaglandin H2 Synthase (Cyclooxygenase)
Cyclin- Dependent Kinase 2
Plasminogen Activator Inhibitor
Stromelysin-1 (Mmp 3)
Lupenone Cyclin- Dependent Kinase 2
Lupeol Nitric Oxide Synthase
Cyclin- Dependent Kinase 2
Stigmasterol Plasminogen Activator Inhibitor
Stromelysin-1 (Mmp 3)
Adp-Ribosylation Factor 6
Fgf Receptor 1
Retinoic Acid Receptor
Nitric Oxide Synthase
Prostaglandin H2 Synthase (Cyclooxygenase)
Estrogen Receptor
Cathepsin B
Ethylcaprylate Tumor Necrosis Factor Receptor
Ethyllaurate Tumor Necrosis Factor Receptor
Ethyloleate Prolyl Endopeptidase
Monolaurin Chymase
Gtp-Binding Protein (Cdc42)
S-Adenosylmethionine Decarboxyla
Nitric Oxide Synthase
Protein Kinase C
Monomyristin Fibroblast Growth Factor 1
Cathepsin B

Among the interaction between the predicted targets and the ingredients of Serenoa

repens that have been mentioned above, many have not been proven or implicated by

experiments. The discrepancy arises for a number of reasons.

It is not expected that exhaustive experiments have been done to determine all

protein targets of a given compound, nor the multiple compounds of one herb. Because

95
 Chapter 4

the protein expression, protein concentration and accessibility vary largely in different

cell system, on which many experimental studies of compound are based, observation

of the molecular events related to the ligand-protein interaction is also difficult. These

targets do not exist or have lower level in these cells. Although the binding energy

suggests that the interaction should be possible, but not exist in these cells. The

discovery of protein profiles can provide useful information to facilitate the study

using our method.

Another discrepancy is caused by the difference of absorption and distribution of

these compounds by different cells or issues. Even if a large doze of one chemical drug

is taken, when it arrives at its protein target, its concentration may become very low

compared with the initial concentration. On the contrary, a chemical with low initial

concentration may easily arrive at its targets to take affect efficiently. In this regards,

when we select the compounds, we chose not only those with great amount in the

extract but also those with little amount. Besides, to determine the potentia l ligands of

one given protein, we take into account of the binding energy. If the binding energy of

a compound was 3 kcal/mol less than the maximum, the binding of the compound is

neglected. Considered that the binding ability is exponential with the binding energy,

this selection role may reasonably pick up the correct interaction.

The capability of the inverse docking approach in identifying potential protein

targets of a small molecule is constrained by the relatively limited number of available

protein 3D structures. This is particularly true for membrane-bond receptor proteins

that are key therapeutic targets for a variety of diseases. Moreover, some of the PDB

96
 Chapter 4

structures may be of little relevance to binding study for a particular molecule. These

include entries containing an incomplete section or a chain, protein mutants that are

structurally different from the corresponding proteins investigated in experiments,

ligand-bound proteins whose conformation is relevant only to a specific set of

compounds and macromolecular complexes unrelated to a particular biological process

studied experimentally. Computer selection of such irrelevant entries may thus

generate ‘false hits’. Anticipated rapid progress instructural genomics [169] is

expected to provide a more diverse set of relevant structures. Knowledge from study of

protein functions also facilitates the selection of relevant structures in determination of

potential protein targets related to a particular cellular or physiological condition.

In our study, 5 alpha reductase, which has been long suggested to be the most

important target in treatment of BPH, is missed because of the lack of 3D structure.

Experiments demonstrate that 5 alpha reductase is a very important enzyme involved

in the proliferatio n of the prostate. Studies show that inhibition of this enzyme could be

a useful method to cure diseases related with the prostate. This problem can be

partially alleviated by rapid progress in high-throughput X-ray crystallography and

NMR of protein structures and in the development of new structural determination

methods [169].

97
 Chapter 4

4.5 Conclusions

Our study provides the potential targets of Serenoa repens in the treatment of BPH,

part of which have been demonstrated by previous experiments to be bound by some

compounds in the extract. Besides these interactions, other predicted bindings between

particular compounds and protein targets have not been proven by experiments.

Moreover, from our data, other important targets may be involved in the pathogenesis

of BPH and can be targeted by Serenoa repens. Thus further studies need to be carried

out to evaluate these protein targets and the role of these compounds. In conclusion, as

a relatively fast-speed and low-cost tool, this software may find application in

systematic study of the molecular mechanism of multiple ingredients of other

medicinal plants.

98
 Chapter 5

Chapter 5: Conclusions

In this work, I develop a new TCM database to provide comprehensive information

about all aspects of TCM including prescription formulae, constituent herbs, known

herbal ingredients and their molecular actions. The therapeutic effects, adverse

reactions, clinical indications and applications are provided at the levels of prescription,

individual herb and herbal ingredient respectively. Traditional terminologies about

medicinal herbs such as tastes, flavors, site of actions, toxicity level, and therapeutic

class are also given.

The study suggests the capability of SVM in recognizing non-effective formulae.

Almost all negative formulae are correctly classified. Especially in Group 2,4,6 and 7,

all of the negative samples are correctly recognized. The study may also have some

helpful hints for herbalist doctors to determine the effectiveness of a TCM formula.

Except Group 4 (formulae contain 7 herbs), the sensitivities of the other groups are

more than 60%. Especially, this method has satisfactory performance for groups in

which the formulae contains 4,6,10 herbs, respectively. In these groups, the

sensitivities are more than 83%. In addition, several potential effective TCM formulae

are derived from hundreds of randomly mixed recipes. These formulae never appear in

previous literature and their effects have never been studied before. It is unclear

99
 Chapter 5

whether these formulae have the therapeutic value. It is expected that the method can

be an alternative approach to study formulae while more and more formulae are under

scientific studies.

The other part of this work is focused on the molecular mechanism study of herbal

medicines by a computer-based method INVDOCK. In this study, I examine the

interaction between the chemical ingredients of a medicinal herb Serenoa repens and a

cavity database to generate the potential targets of these compounds. A total of 25

potential targets are obtained, each of which can be bound by several different

compounds. Among the list of the targets, a number of proteins have been proven by

experiments to be bound by the compounds. Moreover, many protein targets are found

to be involved in cell proliferating, cell cycle, tumor metastasis and other processes

and have higher expression in BPH or prostate cancer tissue, indicating the important

roles in the treatment of BPH. The computational results indicate the feasibility of

INVDOCK in systematic study of the molecular mechanism of multiple ingredients of

Serenoa repens.

100
 References

References

1. Yuan R and Lin Y. Traditional Chinese medicine: an approach to scientific proof

and clinical validation. Pharmacology Therapeutics. 2000, 86:191-198.

2. Chan K. Progress in traditional Chinese medicine. TiPS. 1995, 16:182-187.

3. Porkert M. Chinese Medicine. William Morrow and Company, INC. 1988.

4. Wu DX. The basic principle of traditional Chinese medicine (Chinese). Shanghai

Science and Technology Press. 1994.

5. Ling YK and Yan ZH. Traditional Chine se medicine. Shanghai Science and

Technology Press. 1984.

6. Xu C. Exploration on the characteristics of classification in Huangdi nei jing tai

su. Zhonghua Yi Shi Za Zhi. 1999, 29(2):109-112.

7. Shiang E and Li FP. The Yin-Yang (cold-hot) theory of disease. JAMA. 1971,

217(8):1108.

8. Jia F. Exploration on the medical life of Bian Que, a renowned physician in

ancient China. Zhonghua Yi Shi Za Zhi. 1984, 14(4):214-215.

9. Shang Z. Discussion on the date of appearance of the title Shen nong ben caojing

(Shennong's Herbal Classic). Zhonghua Yi Shi Za Zhi. 1999, 29(3):135-8.

10. Ouyang B. Studies on Shang Han Lun in Ming Dynasty and its influence.

Zhonghua Yi Shi Za Zhi. 1995, 25(2):92-94.

101
 References

11. Song Z. Research on the archaic pronunciation of Ma Bi San for Ma Fei San, a

well-known anaesthetic proposed by Hua Tuo of Han Dynasty. Zhonghua Yi Shi

Za Zhi. 1987, 17(2):76-78.

12. Liu ZZ. A course in English on Chinese materia medica. (15)Li Shi- Zhen--a great

physician and herbalist. Zhong Yao Tong Bao. 1984, 9(4):187-90.

13. Li SZ. Ben Cao Gang Mu. People's Medical Publishing House. 1982.

14. Chen JT. Review: drug therapy in Chinese traditional medicine. Journal of

Clinical Pharmacology. 2000, 40:445-450.

15. Diamond WJ. The clinical practice of complementary, alternative, and western

medicine. CRC Press LLC. 2001.

16. Ang-Lee MK, Moss J and Yuan CS. Herbal medicines and perioperative care.

JAMA 2001, 286:208-216.

17. Eisenberg DM, Davis RB, Ettner SL, Appel S, Wilkey S, Van Rompay M and

Kessler RC. Trends in alternative medicine use in the United States, 1990-1997.

JAMA. 1998, 280:1569-1575.

18. Kam PC and Liew S. Traditional Chinese herbal medicine and anaesthesia.

Anaesthesia. 2002, 57(11):1083-1089.

19. Qu J, Zhang S and Xie R. The Chinese Materia Medica. Shanghai University of

Traditional Chinese Medicine Press. 1990.

20. Xie Z. Prescriptions of traditional chinese medicine. Foreign Languages Press.

2000.

21. Im PO and Yolton RL. Concepts of traditional Oriental medicine. Optometry.

102
 References

2000, 71(10):621-629.

22. Zhang L. Studies of active ingredients of Chinese traditional compound medicine.

Progress in Chemistry. 1999, 11(2):197-199.

23. Yokozawa T, Liu ZW and Chen CP. Protective effects of Glycyrrhizae radix

extract and its compounds in a renal hypoxia (ischemia)-reoxygenation

(reperfusion) model. Phytomedicine. 2000, 6(6):439-445.

24. Niwa K, Hashimoto M, Morishita S, Yokoyama Y, Mori H and Tamaya T.

Preventive effects of Glycyrrhizae radix extract on estrogen-related endometrial

carcinogenesis in mice. Jpn J Cancer Res. 1999, 90(7):726-732.

25. Akao T. Effects of glycyrrhizin and glycyrrhetic acid on the growth, glycyrrhizin

beta-D-glucuronidase and 3 beta- hydroxysteroid dehydrogenase of human

intestinal bacteria. Biol Pharm Bull. 2000, 23(1):104-107.

26. Sun XY, Huang LJ and Liu X, Chemical ingredients of Radix Glycyrrhizae.

ACTA Chinese Medicine and Pharmacology. 1994, 5:40-42.

27. Wang F and Su YH, Pharmacological action and clinical application of Radix

Glycyrrhizae. Lishizhen Medicine and Meteria Medica Research. 2002,

13(5):303-304.

28. Fujisawa Y, Sakamoto M, Matsushita M, Fujita T and Nishioka K. Glycyrrhizin

inhibits the lytic pathway of complement--possible mechanism of its

anti- inflammatory effect on liver cells in viral hepatitis. Microbiol Immunol.

2000, 44(9):799-804.

29. Ou M. Chinese- English manual of common- used in traditional Chinese medicine.

103
 References

Guangdong Science and Technology Press. 1992.

30. Banskota AH, Tezuka Y, Le Tran Q and Kadota S. Chemical constituents and

biological activities of Vietnamese medicinal plants. Curr Top Med Chem. 2003,

3(2):227-248.

31. Singh VK, George CX, Singh N, Agarwal SS and Gupta BM. Combined

treatment of mice with Panax ginseng extract and interferon inducer.

Amplification of host resistance to Semliki forest virus. Planta Med. 1983,

47:234-236.

32. Singh VK, Agarwal SS and Gupta BM, Immunomodulatory activity of Panax

ginseng extract. Planta Med. 1984, 50:462-465.

33. Tianjin Angel Power Allied Pharmaceutical Company. Hu Xing Dan, Tianjin,

China, 1996.

34. Chiatai Qingchunbao Pharmaceutical Co. Ltd. Sheng Mai Ingection. Hangzhou,

China, 1993.

35. Su W, Wu Z and Quan J. Application of fingerprint analysis and computer pattern

interpretation in identification and quality evaluation of traditional Chinese

medicines. Zhong Yao Cai. 2001, 24(4):295-298.

36. Wang LX, Xiao HB, Liang XM and Bi KS. Vectorial angle method for evaluating

the similarity between two chromatographic fingerprints of Chinese herb. Yao

Xue Xue Bao. 2002, 37(9):713-7.

37. Qiao Y, Li P and Wang Y, Journal of Beijing Univ. of TCM (Chinese), 1998,

21(5):17-19.

104
 References

38. Su W, Wu Z, He X and Chen J. Chemical pattern recognition of traditional

Chinese medicine kudingcha (II). Zhong Yao Cai. 1998, 21(4):170-173.

39. Su W, Wu Z, Liang R, Li Q and Huang T. Zhong Yao Cai (Chinese). 1997,

20(11):588-590.

40. Su W, Wu Z and Liang R. World Science and Technology (Chinese). 1999,

3:33-37.

41. Su W. China Journal of Chinese Materia Medica (Chinese). 1997, 22(3):186-188.

42. Chen YZ and Ung CY. Computer automated prediction of potential therapeutic

and toxicity protein targets of bioactive compounds from Chinese medicinal

plants. Am J Chin Med. 2002, 30(1):139-54.

43. Vapnik VN. Statistical Learning Theory. Wiley- interscience, New York, 1995.

44. Cai CZ, Han LY, Ji ZL, Chen X and Chen YZ. SVM-Prot: web-based support

vector machine software for functional classification of a protein from its

primary sequence. Nucleic Acids Res. 2003, 31(13):3692-3697.

45. Levin RM and Das AK. A scientific basis for the therapeutic effects of Pygeum

africanum and Serenoa repens. Urol Res. 2000, 28(3):201-209.

46. Dreikorn K. Phytotherapeutic agents in the treatment of benign prostatic

hyperplasia. Curr Urol Rep. 2000, 1(2):103-109.

47. Wilt TJ, Ishani A, Rutks I and MacDonald R. Phytotherapy for benign prostatic

hyperplasia. Public Health Nutr. 2000, 3(4A):459-472.

48. Gerber GS. Phytotherapy for benign prostatic hyperplasia. Curr Urol Rep. 2002,

3(4):285-291.

105
 References

49. Lowe FC. Phytotherapy in the management of benign prostatic hyperplasia.

Urology. 2001, 58(6 Suppl 1):71-76.

50. Mantle F. The role of alternative medicine in treating postnatal depression.

Complement Ther Nurs Midwifery. 2002, 8(4):197-203.

51. He M, Yan, X, Zhou J and Xie G. Traditional Chinese medicine database and

application on the Web. J Chem Inf Comput Sci. 2001, 41:273-277.

52. Ji YB and Zhang GM. Pharmacological action and application of available

antitumor composition of traditional Chinese medicine (Chinese). Heilongjiang

Science and Technology Press. 1998.

53. Ji YB. Pharmacological action and application of blood-activating and

stasis-eliminating available composition of traditional Chinese medicine

(Chinese). Heilongjiang Science and Technology Press. 1999.

54. Zhao C and Xu YZ, Antitumor herbal medicines and their active compositions.

Traditional Chinese medicine press of China. 1997.

55. Ji YB. Pharmacological action and application of available composition of

traditional Chinese medicine. Heilongjiang Science and Technology Press. 1994.

56. Huang KT. Handbook of composition and pharmacological action of

common-used traditional Chinese medicine. China Pharmaceutical Science and

Technology Press. 1994.

57. Zhang WG, Jia WC, Li SP, Zhang J, Ou YB and Xu XL. Chinese medicated diet.

Shangha i University of Traditional Chinese Medicine Press. 1990.

58. Ou M. Chinese-English manual of common-used prescriptions in traditional

106
 References

Chinese medicine. Guangdong Science and Technology Press. 1991.

59. Liu DL. Manual of common-used prescriptions in traditional Chinese medecine.

People's Military Medical Press. 1996.

60. Li DQ, Shi LH, Gao Y and Jiang JX. Prescription of traditional Chinese medicine.

Shanghai University of Traditional Chinese Medicine Press. 1990.

61. Shen FW. A comprehensive Chinese-Latin- English dictionary of the names of

Chinese herbal medicines. Guangdong World Publishing Co., Ltd. 1998.

62. Zhang EQ. English-Chinese highly efficacious Chinese patent medicines.

Shanghai University of Traditional Chinese Medicine Press. 1998.

63. He R. Xu Lingtai and his medical literatures. Journal of Zhejiang College of

TCM. 2002, 24(6):13-14.

64. de Vel O, Anderson A, Corney M and Mohay G. Mining E-mail content for

author identification forensics. Sigmod Record. 2001, 30,55-64.

65. Kim KI, Jung K, Park SH and Kim HJ. Support vector machine-based text

detection in digital video. Pattern Recogn. 2001, 34:527-529.

66. Drucker H, Wu DH and Vapnik VN. Support vector machine for spam

categorization. IEEE T. Neural Networ. 1999, 10:1048-1054.

67. Vapnik VN. The nature of statistical learning theory, 2nd Ed. Springer, New York,

1999.

68. Tong S and Koller D. Support vector machine active learning with applications to

text classification. J Mach Learn Res. 2001, 2:45-66.

69. Li ZY, Tang SW and Yan SC. Multi-class SVM classifier based on pairwise

107
 References

coupling. LNCS. 2002, 238:321-333.

70. Thubthong N and Kijsirikul B. Support vector machines for Thai phoneme

recognition. Inter J Uncertain Fuzz. 2001, 9:803-813.

71. Gordan M, Kotropoulos C and Pitas I. A temporal network of support vector

machine classifiers for the recognition of visual speech. LNAI. 2002,

230:355-365.

72. Ben-Yacoub S, Abdeljaoued Y and Mayoraz E. Fusion of face and speech data

for person identity verification. IEEE T. Neural Networ. 1999, 10:1065-1074.

73. Wu CY, Liu C and Zhou J. Eyeglasses verification by support vector machine.

LNCS. 2001, 219:1126-1131.

74. Wang YJ, Chua CS and Ho YK. Facial feature detection and face recognition

from 2D and 3D images. Pattern Recogn Lett. 2002, 23:1191-1202.

75. Hsieh JW and Huang YS. Multiple-person tracking system for content analysis.

Inter.J.Pattern Recognition and Artificial Intelligence. 2002, 16:447-462.

76. Papageorgiou C and Poggio T. A trainable system for object detection. Inter J

Comput Vision. 2000, 38:15-33.

77. Karlsen RE, Gorsich DJ and Gerhart GR. Target classification via support vector

machines. Opt Eng. 2000, 39:704-711.

78. Zhao Q, Principe JC, Brennan VL, Xu DX and Wang Z. Synthetic aperture radar

automatic target recognition with three strategies of learning and representation.

Opt Eng. 2000, 39:1230-1244.

79. Gavrishchaka VV and Ganguli SB. Support vector machine as an efficient tool

108
 References

for high-dimensional data processing: application to substorm forecasting. J

Geophysical Res. 2001, 106:29911-29914.

80. Liong SY, Sivapragasam C. Flood stage forecasting with support vector machines.

J Am Water Resour As. 2002, 38:173-186.

81. Fritsche HA. Tumor markers and pattern recognition analysis: a new diagnostic

tool for cancer. J Clin Ligand Assay. 2002, 25:11-15.

82. Bao L and Sun Z. Identifying genes related to drug anticancer mechanisms using

support vector machine. FEBS letters. 2002, 521:109-114.

83. Ramaswamy S, Tamayo P, Rifkin R, Mukherjee S, Yeang CH, Angelo M, Ladd C,

Reich M, Latulippe E, Mesirov JP, Poggio T, Gerald W, Loda M, Lander ES and

Golub TR. Multiclass cancer diagnosis using tumor gene expression signatures.

Proc Natl Acad Sci USA. 2001, 98:15149-15154.

84. Chan K, Lee TW, Sample PA, Goldbaum MH, Weinreb RN and Sejnowski TJ,

Comparison of machine learning and traditional classifiers in glaucoma diagnosis.

IEEE T Biomed Eng. 2002, 49:963-974.

85. Furey TS, Cristianini N, Duffy N, Bednarski DW, Schummer M and Haussler D.

Support vector machine classification and validation of cancer tissue samples

using microarray expression data. Bioinformatics. 2000, 16:906-914.

86. Pavlidis P, Weston J, Cai JS and Noble WS. Learning gene functional

classifications from multiple data types. J Comput Biol. 2002, 9:401-411.

87. Brown MPS, Grundy WN, Lin D, Cristianini N, Sugnet CW, Furey TS, Ares M Jr

and Haussler D. Knowledge-based analysis of microarray gene expression data

109
 References

by using support vector machines. Proc Natl Acad Sci USA. 2000, 97:262-267.

88. Burbidge R, Trotter M, Buxton B and Holden S. Drug design by machine

learning: support vector machines for pharmaceutical data analysis. Comput

Chem. 2001, 26:5-14.

89. Cai YD, Liu XJ, Xu XB and Chou KC. Support Vector Machines for Predicting

HIV Protease Cleavage Sites in Protein. J Comput Chem. 2002, 23:267-274.

90. Wacher VJ, Salphati L and Benet LZ. Active secretion and enterocytic drug

metabolism barriers to drug absorption. Adv Drug Del Rev. 1996, 20:99-112.

91. Ding CH and Dubchak I. Multi-class protein fold recognition using support

vector machines and neural networks, Bioinformatics. 2001, 17:349-358.

92. Baldi P, Brunak S, Chauvin Y, Anderson CAF and Nielsen H. Assessing the

accuracy of prediction algorithms for classification: an overview. Bioinformatics.

2000, 16:412-424.

93. Roulston JE. Screening with Tumor Markers. Mol Biotechnol. 2002, 20:153-162.

94. Buck A. Phytotherapy for the prostate. Br J Urol. 1996, 78:325-336.

95. Shoskes DA. Phytotherapy and other alternative forms of care for the patient

with prostatitis. Curr Urol Rep. 2002, 3(4):330-334.

96. Lowe F and KU J. Phytotherapy in treatment of benign prostatic hyperplasia: a

critical review. Urology. 1996, 48(1):12-20.

97. Wilt T, Ishani A and Mac Donald R. Serenoa repens for benign prostatic

hyperplasia (Cochrane Review). Cochrane Database Syst Rev. 2002,

3:CD001423.

110
 References

98. Debruyne F, Koch G, Boyle P, Da Silva FC, Gillenwater JG, Hamdy FC, Perrin P,

Teillac P, Vela-Navarrete R and Raynaud JP. Comparison of a Phytotherapeutic

Agent (Permixon) with an alpha-Blocker (Tamsulosin) in the Treatment of

Benign Prostatic Hyperplasia: A 1-Year Randomized International Study. Eur

Urol. 2002, 41(5):497-507.

99. Koch E. Extracts from fruits of saw palmetto (Sabal serrulata) and roots of

stinging nettle (Urtica dioica): viable alternatives in the medical treatment of

benign prostatic hyperplasia and associated lower urinary tracts symptoms.

Planta Med. 2001, 67(6):489-500.

100. Boyle P, Robertson C, Lowe F and Roehrborn C. Meta-analysis of clinical trials

of permixon in the treatment of symptomatic benign prostatic hyperplasia.

Urology. 2000, 55(4):533-539.

101. Wilt TJ, Ishani A, Stark G, MacDonald R, Lau J and Mulrow C. Saw palmetto

extracts for treatment of benign prostatic hyperplasia: a systematic review.

JAMA. 1998, 280(18):1604-1609.

102. Gerber GS, Zagaja GP, Bales GT, Chodak GW and Contreras BA. Saw palmetto

(Serenoa repens) in men with lower urinary tract symptoms: effects on

urodynamic parameters and voiding symptoms. Urology. 1998, 51(6):1003-1007.

103. Plosker GL and Brogden RN. Serenoa repens (Permixon). A review of its

pharmacology and therapeutic efficacy in benign prostatic hyperplasia. Drugs

Aging. 1996, 9(5):379-395.

104. Di Silverio F, Flammia GP, Sciarra A, Caponera M, Mauro M, Buscarini M,

111
 References

Tavani M and D'Eramo G. Plant extracts in BPH. Minerva Urol Nefrol. 1993,

45(4):143-149.

105. Di Silverio F, Monti S, Sciarra A, Varasano PA, Martini C, Lanzara S, D'Eramo G,

Di Nicola S and Toscano V. Effects of long-term treatment with Serenoa repens

(Permixon) on the concentrations and regional distribution of androgens and

epidermal growth factor in benign prostatic hyperplasia. Prostate. 1998,

37(2):77-83.

106. Di Silverio F, D'Eramo G, Lubrano C, Flammia GP, Sciarra A, Palma E,

Caponera M and Sciarra F. Evidence that Serenoa repens extract displays an

antiestrogenic activity in prostatic tissue of benign prostatic hypertrophy patients.

Eur Urol. 1992, 21(4):309-314.

107. Paubert-Braquet M, Richardson FO, Servent-Saez N, Gordon WC, Monge MC,

Bazan NG, Authie D and Braquet P. Effect of Serenoa repens extract (Permixon)

on estradiol/testosterone- induced experimental prostate enlargement in the rat.

Pharmacol Res. 1996, 34(3-4):171-179.

108. Carraro JC, Raynaud JP, Koch G, Chisholm GD, Di Silverio F, Teillac P, Da Silva

FC, Cauquil J, Chopin DK, Hamdy FC, Hanus M, Hauri D, Kalinteris A,

Marencak J, Perier A and Perrin P. Comparison of phytotherapy (Permixon) with

finasteride in the treatment of benign prostate hyperplasia: a randomized

international study of 1,098 patients. Prostate. 1996, 29(4):231-240.

109. Bayne CW, Donnelly F, Ross M and Habib FK. Serenoa repens (Permixon): a

5alpha-reductase types I and II inhibitor- new evidence in a coculture model of

112
 References

BPH. Prostate. 1999, 40(4):232-241.

110. Delos S, Iehle C, Martin PM and Raynaud JP. Inhibition of the activity of 'basic'

5 alpha-reductase (type 1) detected in DU 145 cells and expressed in insect cells.

J Steroid Biochem Mol Biol. 1994, 48(4):347-352.

111. Sultan C, Terraza A, Devillier C, Carilla E, Briley M, Loire C and Descomps B.

Inhibition of androgen metabolism and binding by a liposterolic extract of

"Serenoa repens B" in human foreskin fibroblasts. J Steroid Biochem. 1984,

20(1):515-519.

112. Vacherot F, Azzouz M, Gil-Diez-De-Medina S, Colombel M, De La Taille A,

Lefrere Belda MA, Abbou CC, Raynaud JP and Chopin DK. Induction of

apoptosis and inhibition of cell proliferation by the lipido-sterolic extract of

Serenoa repens (LSESr, Permixon in benign prostatic hyperplasia. Prostate. 2000,

45(3):255-266.

113. Paubert-Braquet M, Cous se H, Raynaud JP, Mencia-Huerta JM and Braquet P.

Effect of the lipidosterolic extract of Serenoa repens (Permixon) and its major

components on basic fibroblast growth factor- induced proliferation of cultures of

human prostate biopsies. Eur Urol. 1998, 33(3):340-347.

114. Vacher P, Prevarskaya N, Skryma R, Audy MC, Vacher AM, Odessa MF and

Dufy B. The Lipidosterolic Extract fromSerenoa repens Interferes with Prolactin

Receptor Signal Transduction. J Biomed Sci. 1995, 2(4):357-365.

115. Paubert-Braquet M, Mencia Huerta JM, Cousse H and Braquet P. Effect of the

lipidic lipidosterolic extract of Serenoa repens (Permixon) on the ionophore

113
 References

A23187-stimulated production of leukotriene B4 (LTB4) from human

polymorphonuclear neutrophils. Prostaglandins Leukot Essent Fatty Acids. 1997,

57(3):299-304.

116. Ravenna L, Di Silverio F, Russo MA, Salvatori L, Morgante E, Morrone S,

Cardillo MR, Russo A, Frati L, Gulino A and Petrangeli E. Effects of the

lipidosterolic extract of Serenoa repens (Permixon) on human prostatic cell lines.

Prostate. 1996, 29(4):219-230.

117. Mitropoulos D, Kyroudi A, Zervas A, Papadoukakis S, Giannopoulos A, Kittas C

and Karayannacos P. In vivo effect of the lipido-sterolic extract of Serenoa

repens (Permixon) on mast cell accumulation and glandular epithelium trophism

in the rat prostate. World J Urol. 2002, 19(6):457-561.

118. Shimada H, Tyler V and McLaughlin J. Biologically active acylglycerides from

the berries of Saw-palmetto (Serenoa repens). J Nat Pro. 1997, 60:417-418.

119. Chen YZ and Zhi DG. Ligand-protein inverse docking and its potential use in the

computer search of protein targets of a small molecule. Proteins. 2001,

43(2):217-226.

120. Chen YZ and Ung CY. Prediction of potential toxicity and side effect protein

targets of a small molecule by a ligand-protein inverse docking approach. J Mol

Graph Model. 2001, 20(3):199-218.

121. Kuntz ID, Blaney JM, Oatley SJ, Langridge R and Ferrin TE. A geometric

approach to macromolecule- ligand recognitions. J Mol Biol. 1982, 161:269-288.

122. Richards FM. Areas, volumes, packing and protein structure. Annu Rev Biophys

114
 References

Bioeng. 1977, 6:151-176.

123. Cornell WD, Cieplak P, Bayly CI, Gould IR, Merz KM Jr, Ferguson DM,

Spellmeyer DC, Fox T, Caldwell JW and Kollman PA. A second generation force

field for the simulation of proteins and nucleic acids. J Am Chem Soc. 1995,

117:5179-5197.

124. Wang J, Kollman PA and Kuntz ID. Flexible ligand docking: a multistep strategy

approach. Proteins. 1999, 36:1-19.

125. Baird NC. Simulation of hydrogen bonding in biological systems: ab initio

calculations for NH3-NH3 and NH3-NH4+. Int J Quantum Chem Symp. 1974,

1:49-53.

126. Chen YZ and Prohofsky EW. The role of a minor groove spine of hydration in

stabilizing poly(dA). poly(dT) against fluctuational interbase H-bond disruption

in the premelting temperature regime. Nucleic Acids Res. 1992, 20:415-419.

127. Chen YZ and Prohofsky EW. Premelting base pair opening probability and drug

binding constant of a daunomycin - Poly d(GCAT)-Poly d(ATGC) complex.

Biophys J. 1994, 66:820-826.

128. Carlson HA and McCammon JA. Accommodating protein flexibility in

computational drug design. Mol Pharmacol. 2000, 57:213-218.

129. Carlson HA. Protein flexibility is an important component of structure-based

drug discovery. Curr Pharm Des. 2002, 8(17):1571-8.

130. Ernst E. The risk-benefit profile of commonly used herbal therapies: Ginkgo, St.

John’s Wort, Ginseng, Echinacea, Saw Palmetto, and Kava. Ann Intern Med.

115
 References

2000. 136:42-53.

131. Palin MF, Faguy M, LeHoux JG and Pelletier G. Inhibitory effects of Serenoa

repens on the kinetic of pig prostatic microsomal 5alpha-reductase activity.

Endocrin. 1998, 9(1):65-69.

132. Cristoni A, Pierro FD and Bombardelli E. Botanical derivatives for the prostate.

Fitoterapia. 2000, 71:S21-S28.

133. Chen X, Ji ZL and Chen YZ. TTD: Therapeutic Target Database. Nucleic Acids

Res. 2002, 30(1):412-415.

134. Dvorkin L and Song KY. Herbs for Benign prostatic hyperplasia. Annals of

Pharmacotherapy. 2002, 36:1443-1552.

135. Goldmann WH, Sharma AL, Currier SJ, Johnston PD, Rana A and Sharma CP.

Saw palmetto berry extract inhibits cell growth and Cox-2 expression in prostatic

cancer cells. Cell Biol Int. 2001, 25(11):1117-1124.

136. Boget S, Cereser C, Parvaz P, Leriche A and Revol A. Fibroblast growth factor

receptor 1 (FGFR1) is over-expressed in benign prostatic hyperplasia whereas

FGFR2-IIIc and FGFR3 are not. Eur J Endocrinol. 2001, 145(3):303-310.

137. Deshmukh N, Scotson J, Dodson AR, Smith PH, Ke Y and Foster CS.

Differential expression of acidic and basic fibroblast growth factors in benign

prostatic hyperplasia identified by immunohistochemistry. Br J Urol. 1997,

80(6):869-874.

138. Ozen M, Giri D, Ropiquet F, Mansukhani A and Ittmann M. Role of fibroblast

growth factor receptor signaling in prostate cancer cell survival. J Natl Cancer

116
 References

Inst. 2001, 93(23):1783-1790.

139. Simak R, Capodieci P, Cohen DW, Fair WR, Scher H, Melamed J, Drobnjak M,

Heston WD, Stix U, Steiner G and Cordon-Cardo C. Expression of c-kit and

kit- ligand in benign and malignant prostatic tissues. Histol Histopathol. 2000,

15(2):365-374.

140. Savarese DM, Valinski H, Quesenberry P and Savarese T. Expression and

function of colony-stimulating factors and their receptors in human prostate

carcinoma cell lines. Prostate. 1998, 34(2):80-91.

141. Liu X, Wang L, Lin Y, Teng Q, Zhao C, Hu H and Chi W. Ornithine

decarboxylase activity and its gene expression are increased in benign

hyperplastic prostate. Prostate. 2000, 43(2):83-87.

142. Dunzendorfer U and Knoner M. Therapy with inhibitors of polyamine

biosynthesis in refractory prostatic carcinoma. An experimental and clinical study.

Onkologie. 1985, 8(4):196-200.

143. Haynes JM, Frydenberg M and Majewski H. Testosterone- and phorbol

ester-stimulated proliferation in human cultured prostatic stromal cells. Cell

Signal. 2001, 13(10):703-709.

144. Gradini R, Realacci M, Ginepri A, Naso G, Santangelo C, Cela O, Sale P, Berardi

A, Petrangeli E, Gallucci M, Di Silverio F and Russo MA. Nitric oxide synthases

in normal and benign hyperplastic human prostate: immunohistochemistry and

molecular biology. J Pathol. 1999, 189(2):224-229.

145. Pasquali D, Thaller C and Eichele G. Abnormal level of retinoic acid in prostate

117
 References

cancer tissues. J Clin Endocrinol Metab. 1996, 81(6):2186-2191.

146. He XY, Merz G, Yang YZ, Pullakart R, Mehta P, Schulz H and Yang SY. Function

of human brain short chain L-3-hydroxyacyl coenzyme A dehydrogenase in

androgen metabolism. Biochim Biophys Acta. 2000, 1484(2-3):267-277.

147. Mohan RR, Challa A, Gupta S, Bostwick DG, Ahmad N, Agarwal R, Marengo

SR, Amini SB, Paras F, MacLennan GT, Resnick MI and Mukhtar H.

Overexpression of ornithine decarboxylase in prostate cancer and prostatic fluid

in humans. Clin Cancer Res. 1999, 5(1):143-147.

148. Rosenblum ER, Stauber RE, Van Thiel DH, Campbell IM and Gavaler JS.

Assessment of the estrogenic activity of phytoestrogens isolated from bourbon

and beer. Alcohol Clin Exp Res. 1993, 17(6):1207-1209.

149. Carnero A. Targeting the cell cycle for cancer therapy. Br J Cancer. 2002,

87(2):129-133.

150. Garner-Hamrick PA and Fisher C. Antisense phosphorothioate oligonucleotides

specifically down-regulate cdc25B causing S-phase delay and persistent

antiproliferative effects. Int J Cancer. 1998, 76(5):720-728.

151. Draetta G and Eckstein J. Cdc25 protein phosphatases in cell proliferation.

Biochim Biophys Acta. 1997, 1332(2):M53-M63.

152. Aznar S and Lacal JC. Rho signals to cell growth and apoptosis. Cancer Lett.

2001, 165(1):1-10.

153. Erickson JW and Cerione RA. Multiple roles for Cdc42 in cell regulation. Curr

Opin Cell Biol. 2001, 13(2):153-157.

118
 References

154. Murant SJ, Handley J, Stower M, Reid N, Cussenot O and Maitland NJ.

Co-ordinated changes in expression of cell adhesion molecules in prostate cancer.

Eur J Cancer. 1997, 33(2):263-271.

155. Arenas MI, Romo E, Royuela M, Fraile B and Paniagua R. E-, N- and P-cadherin,

and alpha-, beta- and gamma-catenin protein expression in normal, hyperplastic

and carcinomatous human prostate. Histochem J. 2000, 32(11):659-667.

156. Szpaderska AM and Frankfater A. An Intracellular Form of Cathepsin B

Contributes to Invasiveness in Cancer. Cancer Res. 2001, 61(8):3493-3500.

157. Goossens F, De Meester I, Vanhoof G and Scharpe S. Distribution of prolyl

oligopeptidase in human peripheral tissues and body fluids. Eur J Clin Chem Clin

Biochem. 1996, 34(1):17-22.

158. Jung K, Nowak L, Lein M, Priem F, Schnorr D and Loening SA. Matrix

metalloproteinases 1 and 3, tissue inhibitor of metalloproteinase-1 and the

complex of metalloproteinase-1/tissue inhibitor in plasma of patients with

prostate cancer. Int J Cancer. 1997, 74(2):220-223.

159. Turner CE and Brown MC. Cell motility: ARNO and ARF6 at the cutting edge.

Curr Biol. 2001, 11(21):R875-R877

160. Neumann U SNaGM. Inhibition of Human Chymase by

2-Amino-3,1-benzoxazin-4-ones. Bioorganic and Medicinal chemistry. 2002,

9:947-954.

161. de Miguel MP, Royuela M, Bethencourt FR, Santamaria L, Fraile B and

Paniagua R. Immunoexpression of tumour necrosis factor-alpha and its receptors

119
 References

1 and 2 correlates with proliferation/apoptosis equilibrium in normal, hyperplasic

and carcinomatous human prostate. Cytokine. 2000, 12(5):535-538.

162. Kwaan HC, Wang J, Svoboda K and Declerck PJ. Plasminogen activator

inhibitor 1 may promote tumour growth through inhibition of apoptosis. Br J

Cancer. 2000, 82(10):1702-1708.

163. Awad AB, Williams H and Fink CS. Phytosterols reduce in vitro metastatic

ability of MDA-MB-231 human breast cancer cells. Nutr Cancer. 2001,

40(2):157-164.

164. Connolly JM, Coleman M and Rose DP. Effects of dietary fatty acids on DU145

human prostate cancer cell growth in athymic nude mice. Nutr Cancer. 1997,

29(2):114-119.

165. Senzaki H, Iwamoto S, Ogura E, Kiyozuka Y, Arita S, Kurebayashi J, Takada H,

Hioki K and Tsubura A. Dietary effects of fatty acids on growth and metastasis of

KPL-1 human breast cancer cells in vivo and in vitro. Anticancer Res. 1998,

18(3A):1621-1627.

166. Iwamoto S, Senzaki H, Kiyozuka Y, Ogura E, Takada H, Hioki K and Tsubura A.

Effects of fatty acids on liver metastasis of ACL-15 rat colon cancer cells. Nutr

Cancer. 1998, 31(2):143-150.

167. Awad AB, Fink CS, Williams H and Kim U. In vitro and in vivo (SCID mice)

effects of phytosterols on the growth and dissemination of human prostate cancer

PC-3 cells. Eur J Cancer Prev. 2002, 10(6):507-513.

168. Awad AB, Downie A, Fink CS and Kim U. Dietary phytosterol inhibits the

120
 References

growth and metastasis of MDA-MB-231 human breast cancer cells grown in

SCID mice. Anticancer Res. 2000, 20(2A):821-824.

169. Sali A. 100,000 protein structures for the biologist. Nat Struct Biol. 1998,

5(12):1029-1032.

121